Enterprise-level SSD resource scheduling system based on storage virtualization

Abstract

This invention discloses an enterprise-level SSD resource scheduling system based on storage virtualization, relating to the fields of storage scheduling and data processing technology. It includes a virtual resource modeling module that performs aggregation and association processing, service level binding processing, and access feature organization processing on tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics to generate virtual resource status data. In this invention, the virtual resource modeling module performs aggregation and association, service level binding, and access feature organization; the media status awareness module performs disk-level correspondence organization, time-sequence organization, and media pressure merging; and the dual-state mapping construction module performs resource instance correspondence processing, media bearer association processing, and gray relational analysis algorithm processing based on dynamic time warping alignment. This achieves continuous mapping between virtual resource instances and media operating states, explicit bearer relationships, and unified cross-layer state expression.

Description

Technical Field

[0001] This invention relates to the field of storage scheduling and data processing technology, and in particular to an enterprise-level SSD resource scheduling system based on storage virtualization. Background Technology

[0002] With the continued development of multi-tenant services and virtualization deployments, enterprise-grade SSD resource scheduling systems based on storage virtualization are increasingly being used to organize and process tenant-side request information and media-side operational information. This further forms a chain of processes including dual-state mapping, scheduling budgets, partial execution, and feedback updates, enabling dynamic adjustment of namespace binding, virtual function allocation, queue delivery, and bandwidth usage. However, in application environments where service levels, resource contention, and media operational states are continuously coupled, existing systems are gradually revealing technical challenges that require further resolution.

[0003] Existing enterprise-level SSD resource scheduling solutions in storage virtualization scenarios typically handle tenant-side request characteristics and media-side operating status separately, making it difficult to establish a stable correspondence between virtual resource status and media operating status. This leads to difficulties in timely identification of resource contention, overlapping media pressure, and concentrated load. At the same time, the scheduling process lacks a continuous coordination mechanism based on business level and resource occupancy status, which can easily cause asynchrony between namespace binding, virtual function allocation, queue delivery, and bandwidth usage adjustments, thereby affecting the stability of execution results and the timeliness of feedback adjustments. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an enterprise-level SSD resource scheduling system based on storage virtualization.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an enterprise-level SSD resource scheduling system based on storage virtualization, comprising: a virtual resource modeling module, which performs aggregation and association processing, service level binding processing, and access feature sorting processing on tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics to generate virtual resource status data; a media status awareness module, which performs disk-level correspondence sorting processing, time-sequence organization processing, and media pressure merging processing on SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information to generate media operating status data; and a dual-state mapping construction module, which performs resource instance correspondence processing and media bearer association processing on virtual resource status data and media operating status data, and performs resource instance correspondence processing and media bearer association processing on the data. The content after the associated processing is processed by a gray relational analysis algorithm based on dynamic time warping and alignment, and conflict relationship marking, generating bi-state mapping data. The scheduling budget generation module processes the bi-state mapping data by performing business level association processing, resource occupancy assessment processing, Jaya algorithm processing based on simulated annealing acceptance mechanism processing, and scheduling allocation generation processing, generating scheduling budget data. The local execution control module processes the scheduling budget data by performing namespace binding adjustment processing, virtual function allocation adjustment processing, queue delivery rhythm adjustment processing, and bandwidth occupancy rhythm adjustment processing, generating local execution results. The feedback update module processes the local execution results by performing business completion delay aggregation processing, throughput change sorting processing, fair occupancy comparison processing, and media lifetime change association processing, generating service assurance feedback data, and outputting the service assurance feedback data to the scheduling budget generation module.

[0006] As a further description of the above technical solution: The virtual resource modeling module receives tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics. It then aggregates and organizes these identifiers to form a resource identifier aggregation result. Based on this result, it performs business level binding processing on the tenant identifier, service volume identifier, namespace identifier, and virtual function identifier corresponding to each virtual resource instance, establishing a correspondence between each virtual resource instance and its corresponding business level constraints, thus forming a business level binding result. Finally, based on the resource identifier aggregation result and the business level binding result, it performs access characteristic processing on the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance, establishing a correspondence between the request load, request arrival status, and queue occupancy status of each virtual resource instance, generating virtual resource status data.

[0007] As a further description of the above technical solution: The media status awareness module receives SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information. It then performs on-disk organization of these information in chronological order, generating a media status organization result. Based on this result, it performs chronological organization processing on the SSD channel occupancy and media unit load information for each SSD, establishing a sequential correspondence between changes in SSD channel occupancy and media unit load at different times, generating an occupancy / load organization result. Finally, based on the media status organization and occupancy / load organization results, it performs media pressure aggregation processing on the garbage collection activity, wear change, and write amplification change information, establishing a pressure correspondence between the garbage collection changes, wear changes, and write amplification changes for each SSD and the changes in SSD channel occupancy and media unit load, generating media operating status data.

[0008] As a further description of the above technical solution: The dual-state mapping construction module receives tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics from virtual resource status data, and receives SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information from media operation status data. It performs resource instance mapping between the virtual resource status data and the media operation status data to form a resource instance mapping result. Based on the resource instance mapping result, it maps the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance to the SSD channel occupancy information, media unit load information, garbage collection activity information, and wear and tear information in the resource instance mapping result. Information and write amplification change information are processed to form media bearer association results. Based on the media bearer association results, the load expression content corresponding to each virtual resource instance and the load expression content in the media operation status data are processed by a gray relational analysis algorithm based on dynamic time warping alignment to form load coupling processing results. Based on the load coupling processing results, the resource competition relationship, media pressure overlap relationship and load concentration relationship between virtual resource instances, between media operation states and between virtual resource instances and media operation states are processed to form conflict relationship marking results. Based on the load coupling processing results and conflict relationship marking results, corresponding integration processing is performed to generate bi-state mapping data.

[0009] As a further description of the above technical solution: The grey relational analysis algorithm based on dynamic time warping alignment processes the following steps: Based on the media bearer association results, it receives the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance, as well as the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information after bearer association with each virtual resource instance. It then organizes the IO queue occupancy information and access request characteristics into a virtual resource load sequence, and organizes the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information into a media operating load sequence. Finally, it performs dimensionless processing on the virtual resource load sequence and the media operating load sequence, ensuring that the load changes in the virtual resource load sequence align with the load changes in the media operating load sequence. The process involves several steps: first, converting the virtual resource load sequence to a unified comparison scale; second, performing dynamic time warping and alignment on the dimensionless virtual resource load sequence and media operation load sequence, aligning them temporally along the path with the minimum cumulative difference to form a time alignment result; third, constructing a difference sequence on the virtual resource load sequence and media operation load sequence to ensure that the changes in the virtual resource load and media operation load are matched item by item to form a difference sequence; fourth, calculating the grey relational coefficient based on the difference sequence and then calculating the grey relational degree based on the grey relational coefficient to establish a correlation degree between each virtual resource instance and each media operation state; and finally, performing load coupling consolidation based on the correlation degree result to form a load coupling consolidation result.

[0010] As a further description of the above technical solution: The scheduling budget generation module receives the load coupling sorting results and conflict relationship marking results from the bi-state mapping data, and performs corresponding sorting according to virtual resource instances to form the basic budget generation results. Based on the basic budget generation results, it performs business level association processing on each virtual resource instance to establish a correspondence between each virtual resource instance and its corresponding business level constraints, generating business level association results. Based on the basic budget generation results, it performs resource occupancy assessment processing on the load coupling sorting results and conflict relationship marking results corresponding to each virtual resource instance to establish an assessment correspondence between the resource contention level, media pressure level, and load concentration level corresponding to each virtual resource instance, generating resource occupancy assessment results. Based on the business level association results and resource occupancy assessment results, it performs Jaya algorithm processing based on simulated annealing acceptance mechanism to generate media pressure coordination results. Based on the media pressure coordination results, it performs scheduling allocation generation processing on the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth occupancy rhythm adjustment requirements corresponding to each virtual resource instance, generating scheduling budget data.

[0011] As a further description of the above technical solution: The Jaya algorithm based on the simulated annealing acceptance mechanism includes: receiving service level association results and resource occupancy assessment results, and combining and organizing these results to generate a candidate coordination solution set; based on the candidate coordination solution set, identifying the current best and worst candidate coordination solutions, and performing Jaya update processing on the remaining candidate coordination solutions according to the current best and worst candidate coordination solutions to generate an updated candidate coordination solution set; performing fitness comparison processing on each updated candidate coordination solution in the updated candidate coordination solution set and its corresponding original candidate coordination solution, and calculating the fitness difference between the updated candidate coordination solution and the original candidate coordination solution; when the updated candidate coordination solution is better than the corresponding original candidate coordination solution, the algorithm updates the candidate coordination solution set. When selecting candidate coordination solutions, the updated candidate coordination solutions are retained. Temperature decay update processing is performed based on the current iteration round to form the current temperature. When an updated candidate coordination solution is inferior to the corresponding original candidate coordination solution, simulated annealing acceptance judgment is performed based on the current temperature and fitness difference. Updated candidate coordination solutions that meet the acceptance conditions are retained, while those that do not are discarded. Based on the retained candidate coordination solution set, the current best candidate coordination solution identification, current worst candidate coordination solution identification, Jaya update processing, fitness comparison processing, and simulated annealing acceptance judgment are repeatedly performed until the iteration termination condition is met. The current best candidate coordination solution retained when the iteration termination condition is met is determined as the medium pressure coordination result.

[0012] As a further description of the above technical solution: The scheduling allocation generation process includes: receiving the media pressure coordination results and organizing them according to virtual resource instances to form the basic allocation generation results; based on the basic allocation generation results, performing namespace binding adjustment requirement generation processing on the media pressure coordination results corresponding to each virtual resource instance, so that changes in namespace binding for each virtual resource instance form adjustment correspondences; based on the basic allocation generation results, performing virtual function allocation adjustment requirement generation processing on the media pressure coordination results corresponding to each virtual resource instance, so that changes in virtual function allocation for each virtual resource instance form adjustment correspondences; based on the basic allocation generation results, performing queue delivery rhythm adjustment requirement generation processing on the media pressure coordination results corresponding to each virtual resource instance, so that changes in queue delivery for each virtual resource instance form adjustment correspondences; based on the basic allocation generation results, performing bandwidth occupancy rhythm adjustment requirement generation processing on the media pressure coordination results corresponding to each virtual resource instance, so that changes in bandwidth occupancy for each virtual resource instance form adjustment correspondences; and performing corresponding integration processing on the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth occupancy rhythm adjustment requirements to generate scheduling budget data.

[0013] As a further description of the above technical solution: The local execution control module receives namespace binding adjustment requests, virtual function allocation adjustment requests, queue delivery rhythm adjustment requests, and bandwidth usage rhythm adjustment requests from the scheduling budget data. It then organizes these requests according to virtual resource instances, forming the basic execution control results. Based on these results, it performs namespace binding adjustment processing on the namespace binding adjustment requests for each virtual resource instance, resulting in adjusted namespace binding relationships. It also performs virtual function allocation adjustment processing on the virtual function allocation requests for each virtual resource instance, resulting in adjusted virtual function allocation relationships. Furthermore, it performs queue delivery rhythm adjustment processing on the queue delivery rhythm adjustment requests for each virtual resource instance, resulting in adjusted queue delivery relationships. Finally, it performs bandwidth usage rhythm adjustment processing on the bandwidth usage rhythm adjustment requests for each virtual resource instance, resulting in adjusted bandwidth usage relationships. Finally, it performs corresponding integration processing on the namespace binding adjustment results, virtual function allocation adjustment results, queue delivery rhythm adjustment results, and bandwidth usage rhythm adjustment results to generate local execution results.

[0014] As a further description of the above technical solution: The feedback update module receives partial execution results and organizes them according to virtual resource instances to form basic feedback update results. Based on the basic feedback update results, it performs business completion latency aggregation processing on the execution results corresponding to each virtual resource instance, so that the business completion latency corresponding to each virtual resource instance forms an aggregation result. Based on the basic feedback update results, it performs throughput change aggregation processing on the execution results corresponding to each virtual resource instance, so that the throughput change corresponding to each virtual resource instance forms an aggregation result. Based on the basic feedback update results, it performs fair occupancy comparison processing on the execution results corresponding to each virtual resource instance, so that the resource occupancy differences corresponding to each virtual resource instance form a comparison result. Based on the basic feedback update results, it performs media lifetime change correlation processing on the execution results corresponding to each virtual resource instance, so that the lifetime changes corresponding to each virtual resource instance form a correlation result. Based on the business completion latency aggregation result, throughput change aggregation result, fair occupancy comparison result, and media lifetime change correlation result, it performs corresponding integration processing to generate service assurance feedback data, and outputs the service assurance feedback data to the scheduling budget generation module.

[0015] The present invention has the following beneficial effects: 1. In this invention, the virtual resource modeling module first aggregates and associates tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics, binds them to service levels, and organizes access characteristics. Then, the media status awareness module organizes SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information within the disk, organizes them in time order, and merges them with media pressure. Finally, the dual-state mapping construction module performs resource instance correspondence processing, media bearing association processing, and gray relational analysis algorithm processing based on dynamic time regularization alignment. This solves the technical defects of scattered processing of tenant-side request characteristics and media-side operating status in storage virtualization scenarios, and the difficulty in establishing a stable correspondence between virtual resource status and media operating status. It realizes continuous mapping between virtual resource instances and media operating status, explicit bearing relationship, and unified expression of cross-layer status.

[0016] 2. In this invention, the load coupling sorting results and conflict relationship marking results in the dual-state mapping data are combined with the business level constraints by the scheduling budget generation module and incorporated into the business level association processing, resource occupancy assessment processing, and Jaya algorithm processing based on the simulated annealing acceptance mechanism. This solves the technical defects of the scheduling process lacking a continuous coordination mechanism based on business level and resource occupancy status, and the difficulty of different resource instances forming a unified coordination decision oriented towards media pressure. It realizes the joint optimization of business level constraints, resource competition degree, media pressure degree and load concentration degree, and the stable generation of media pressure coordination results. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Figure 1This invention provides an embodiment of an enterprise-level SSD resource scheduling system based on storage virtualization, comprising: a virtual resource modeling module, which performs aggregation and association processing, service level binding processing, and access feature sorting processing on tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics to generate virtual resource status data; a media status awareness module, which performs disk-level correspondence sorting processing, time-sequence organization processing, and media pressure merging processing on SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information to generate media operating status data; and a dual-state mapping construction module, which performs resource instance correspondence processing and media bearer association processing on the virtual resource status data and media operating status data, and performs resource instance correspondence processing and media bearer association processing on the data. The processed content undergoes gray relational analysis based on dynamic time warping and alignment, along with conflict relationship marking, to generate bi-state mapping data. The scheduling budget generation module performs business level association processing, resource occupancy assessment, Jaya algorithm processing based on simulated annealing acceptance mechanism, and scheduling allocation generation on the bi-state mapping data to generate scheduling budget data. The local execution control module performs namespace binding adjustment, virtual function allocation adjustment, queue delivery rhythm adjustment, and bandwidth occupancy rhythm adjustment on the scheduling budget data to generate partial execution results. The feedback update module performs business completion delay aggregation processing, throughput change processing, fair occupancy comparison processing, and media lifetime change association processing on the partial execution results to generate service assurance feedback data, which is then output to the scheduling budget generation module.

[0020] The virtual resource modeling module receives tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics. It then aggregates and organizes these identifiers to form a resource identifier aggregation result. Based on this result, it performs business level binding processing on the tenant identifier, service volume identifier, namespace identifier, and virtual function identifier corresponding to each virtual resource instance, establishing a correspondence between each virtual resource instance and its corresponding business level constraints, thus forming a business level binding result. Finally, based on the resource identifier aggregation result and the business level binding result, it performs access characteristic processing on the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance, establishing a correspondence between the request load, request arrival status, and queue occupancy status of each virtual resource instance, generating virtual resource status data.

[0021] The media status awareness module receives SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information. It then performs on-disk organization of these information in chronological order, generating a media status organization result. Based on this result, it performs chronological organization processing on the SSD channel occupancy and media unit load information for each SSD, establishing a sequential correspondence between changes in SSD channel occupancy and media unit load at different times, generating an occupancy / load organization result. Finally, based on the media status organization and occupancy / load organization results, it performs media pressure aggregation processing on the garbage collection activity, wear change, and write amplification change information, establishing a pressure correspondence between the garbage collection changes, wear changes, and write amplification changes for each SSD and the changes in SSD channel occupancy and media unit load, generating media operating status data.

[0022] The dual-state mapping construction module receives tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics from virtual resource status data, and receives SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information from media operation status data. It performs resource instance mapping between the virtual resource status data and the media operation status data to form a resource instance mapping result. Based on the resource instance mapping result, it maps the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance to the SSD channel occupancy information, media unit load information, garbage collection activity information, and wear and tear information in the resource instance mapping result. Information and write amplification change information are processed to form media bearer association results. Based on the media bearer association results, the load expression content corresponding to each virtual resource instance and the load expression content in the media operation status data are processed by a gray relational analysis algorithm based on dynamic time warping alignment to form load coupling processing results. Based on the load coupling processing results, the resource competition relationship, media pressure overlap relationship and load concentration relationship between virtual resource instances, between media operation states and between virtual resource instances and media operation states are processed to form conflict relationship marking results. Based on the load coupling processing results and conflict relationship marking results, corresponding integration processing is performed to generate bi-state mapping data.

[0023] The scheduling budget generation module receives the load coupling sorting results and conflict relationship marking results from the bi-state mapping data, and performs corresponding sorting according to virtual resource instances to form the basic budget generation results. Based on the basic budget generation results, it performs business level association processing on each virtual resource instance to establish a correspondence between each virtual resource instance and its corresponding business level constraints, generating business level association results. Based on the basic budget generation results, it performs resource occupancy assessment processing on the load coupling sorting results and conflict relationship marking results corresponding to each virtual resource instance to establish an assessment correspondence between the resource contention level, media pressure level, and load concentration level corresponding to each virtual resource instance, generating resource occupancy assessment results. Based on the business level association results and resource occupancy assessment results, it performs Jaya algorithm processing based on simulated annealing acceptance mechanism to generate media pressure coordination results. Based on the media pressure coordination results, it performs scheduling allocation generation processing on the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth occupancy rhythm adjustment requirements corresponding to each virtual resource instance, generating scheduling budget data.

[0024] The local execution control module receives namespace binding adjustment requests, virtual function allocation adjustment requests, queue delivery rhythm adjustment requests, and bandwidth usage rhythm adjustment requests from the scheduling budget data. It then organizes these requests according to virtual resource instances, forming the basic execution control results. Based on these results, it performs namespace binding adjustment processing on the namespace binding adjustment requests for each virtual resource instance, resulting in adjusted namespace binding relationships. It also performs virtual function allocation adjustment processing on the virtual function allocation requests for each virtual resource instance, resulting in adjusted virtual function allocation relationships. Furthermore, it performs queue delivery rhythm adjustment processing on the queue delivery rhythm adjustment requests for each virtual resource instance, resulting in adjusted queue delivery relationships. Finally, it performs bandwidth usage rhythm adjustment processing on the bandwidth usage rhythm adjustment requests for each virtual resource instance, resulting in adjusted bandwidth usage relationships. Finally, it performs corresponding integration processing on the namespace binding adjustment results, virtual function allocation adjustment results, queue delivery rhythm adjustment results, and bandwidth usage rhythm adjustment results to generate local execution results.

[0025] The feedback update module receives partial execution results and organizes them according to virtual resource instances to form basic feedback update results. Based on the basic feedback update results, it performs business completion latency aggregation processing on the execution results corresponding to each virtual resource instance, so that the business completion latency corresponding to each virtual resource instance forms an aggregation result. Based on the basic feedback update results, it performs throughput change aggregation processing on the execution results corresponding to each virtual resource instance, so that the throughput change corresponding to each virtual resource instance forms an aggregation result. Based on the basic feedback update results, it performs fair occupancy comparison processing on the execution results corresponding to each virtual resource instance, so that the resource occupancy differences corresponding to each virtual resource instance form a comparison result. Based on the basic feedback update results, it performs media lifetime change correlation processing on the execution results corresponding to each virtual resource instance, so that the lifetime changes corresponding to each virtual resource instance form a correlation result. Based on the business completion latency aggregation result, throughput change aggregation result, fair occupancy comparison result, and media lifetime change correlation result, it performs corresponding integration processing to generate service assurance feedback data, and outputs the service assurance feedback data to the scheduling budget generation module.

[0026] In this embodiment, the virtual resource modeling module receives tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics. It then performs aggregation and mapping on the received tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics. First, it retrieves the tenant identifiers corresponding to the same virtual resource instance item by item. Then, it retrieves the service volume identifiers corresponding to the same tenant identifier item by item, ensuring a one-to-one correspondence between the service volume identifiers and tenant identifiers. After the tenant identifiers and service volume identifiers are mapped, it continues to retrieve the namespace identifiers corresponding to the same virtual resource instance, ensuring that the namespace identifiers are consistent with the previously mapped tenant identifiers and service volume identifiers. The mapping relationship under resource instances is established. After the namespace identifier is mapped, the virtual function identifier corresponding to the same virtual resource instance is retrieved, and the virtual function identifier is continuously mapped to the tenant identifier, service volume identifier, and namespace identifier in the same aggregation link. Then, the IO queue occupancy information and access request characteristics are retrieved respectively, and the IO queue occupancy information and access request characteristics are mapped to the tenant identifier, service volume identifier, namespace identifier, and virtual function identifier that have been sorted. After all mapping relationships are established, the tenant identifier, service volume identifier, namespace identifier, virtual function identifier, IO queue occupancy information, and access request characteristics are merged and written according to the virtual resource instance to form the resource identifier aggregation result.

[0027] After the resource identifier aggregation results are formed, business level binding processing is performed on the tenant identifier, business volume identifier, namespace identifier, and virtual function identifier corresponding to each virtual resource instance. First, the tenant identifier corresponding to the same virtual resource instance in the resource identifier aggregation results is read, and this tenant identifier is used as the starting point for reading business level binding. After the tenant identifier is read, the business volume identifier, namespace identifier, and virtual function identifier that have been aggregated with the tenant identifier are read, and the continuous correspondence of these four items under the same virtual resource instance is maintained. Then, the corresponding business level constraints are introduced one by one, and the business level constraints are bound to the tenant identifier, business volume identifier, namespace identifier, and virtual function identifier respectively. After the business level constraints are matched, the tenant identifier, business volume identifier, namespace identifier, virtual function identifier, and business level constraints under the same virtual resource instance are written into the same result chain in the original corresponding order to form the business level binding result.

[0028] After the business level binding results are formed, access feature processing is performed on the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance. First, the IO queue occupancy information corresponding to the same virtual resource instance is read from the resource identifier aggregation results, and the original correspondence between the IO queue occupancy information and the tenant identifier, business volume identifier, namespace identifier, and virtual function identifier corresponding to the virtual resource instance is maintained. Then, the access request characteristics corresponding to the same virtual resource instance are read, and the access request characteristics and IO queue occupancy information are made to correspond to each other under the same virtual resource instance. On this basis, each access content in the access request characteristics is processed item by item, so that the access request characteristics under the same virtual resource instance are continuously expanded according to the correspondence. Then, the IO queue occupancy information and the processed access request characteristics are mapped to the business level binding results, so that the request load, request arrival, and queue occupancy corresponding to the same virtual resource instance are made to form a unified correspondence. After the correspondence is completed, the request load, request arrival, and queue occupancy are processed and written together with the tenant identifier, business volume identifier, namespace identifier, virtual function identifier, and business level constraints corresponding to the virtual resource instance to generate virtual resource status data.

[0029] The media status awareness module receives SSD channel occupancy information, media cell load information, garbage collection activity information, wear and tear information, and write amplification change information. It then performs on-disk organization on these data in chronological order. First, it reads the SSD channel occupancy information corresponding to the same SSD item by item in chronological order. Next, it reads the media cell load information corresponding to the same SSD item by item in the same chronological order, ensuring that the media cell load information and SSD channel occupancy information correspond at the same time point. After the SSD channel occupancy information and media cell load information are time-correlated, it continues to read the garbage collection activity information corresponding to the same SSD in chronological order, ensuring that the garbage collection activity information... The time correspondence is kept consistent with the previously completed SSD channel occupancy information and media unit load information. Then, wear change information is read sequentially, ensuring that it maintains an intra-disk correspondence with the SSD channel occupancy information, media unit load information, and garbage collection activity information at the same time point. Next, write amplification change information is read sequentially, ensuring a continuous intra-disk correspondence between write amplification change information and the SSD channel occupancy information, media unit load information, garbage collection activity information, and wear change information for the same SSD. After all the above information has been time-sequentially correlated, the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information are merged and organized separately for each SSD to form a media status organization result.

[0030] After the media status is organized, the SSD channel occupancy information and media unit load information corresponding to each SSD are processed in chronological order. First, the SSD channel occupancy information of the same SSD at different times is read and arranged in chronological order. Then, the media unit load information of the same SSD at different times is read and arranged in the same chronological order as the SSD channel occupancy information. After both types of information are arranged, the changes in SSD channel occupancy at adjacent time positions are linked one by one, and the changes in media unit load at adjacent time positions are linked one by one, so that the changes in SSD channel occupancy and media unit load at different times form a continuous sequential correspondence. After the sequential correspondence is established, the changes in SSD channel occupancy and media unit load at each time position under the same SSD are organized and written together to form the occupancy and load organization result.

[0031] After the occupancy organization results are generated, media pressure merging processing is performed on the garbage collection activity information, wear change information, and write amplification change information. First, the garbage collection activity information corresponding to the same SSD in the media status organization results is read, and the garbage collection activity information is aligned with the SSD channel occupancy change and media unit load change in the occupancy organization results at the same time point. Then, the wear change information corresponding to the same SSD is read, and the wear change information is aligned with the garbage collection activity information, SSD channel occupancy change, and media unit load change. Finally, the write amplification change information corresponding to the same SSD is read, and the write amplification change information is aligned with the garbage collection activity information, SSD channel occupancy change, and media unit load change. Information on garbage collection activities, wear and tear changes, SSD channel occupancy changes, and media unit load changes are uniformly correlated. After all five types of information are correlated, garbage collection changes, wear and tear changes, and write amplification changes are merged into SSD channel occupancy changes and media unit load changes, so that the garbage collection changes, wear and tear changes, and write amplification changes for each SSD are correlated with the SSD channel occupancy changes and media unit load changes under pressure. After the pressure correlation is established, the garbage collection changes, wear and tear changes, write amplification changes, SSD channel occupancy changes, and media unit load changes that have been merged under each SSD are jointly organized and written to generate media operating status data.

[0032] In this embodiment, the dual-state mapping construction module receives tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics from the virtual resource status data, and receives SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information from the media operation status data. When processing resource instance correspondences, it first reads the tenant identifiers in the virtual resource status data item by item for each virtual resource instance, then reads the service volume identifiers, namespace identifiers, and virtual function identifiers item by item for the same virtual resource instance, ensuring a continuous correspondence between the tenant identifiers, service volume identifiers, namespace identifiers, and virtual function identifiers under the same virtual resource instance. After completing the identifier reading, it continues to read the IO queue occupancy information and access request characteristics corresponding to the same virtual resource instance item by item, ensuring a continuous correspondence between the tenant identifiers, service volume identifiers, namespace identifiers, and virtual function identifiers under the same virtual resource instance. The queue occupancy information and access request characteristics are maintained in correspondence with the tenant identifier, service volume identifier, namespace identifier, and virtual function identifier under the same virtual resource instance. Subsequently, the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information in the media operation status data are read in the same correspondence order as the virtual resource instance. The SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information are then matched with the previously read tenant identifier, service volume identifier, namespace identifier, virtual function identifier, IO queue occupancy information, and access request characteristics to form a correspondence under the same virtual resource instance. After all correspondences are established, the virtual resource status data and media operation status data are merged and organized according to the virtual resource instance to form the resource instance correspondence result.

[0033] After the resource instance mapping results are generated, the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance are correlated with the SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information in the resource instance mapping results. First, the IO queue occupancy information corresponding to the same virtual resource instance is retrieved item by item, ensuring a continuous correspondence between this IO queue occupancy information and the corresponding SSD channel occupancy information and media unit load information. After the IO queue occupancy information is mapped, the access request characteristics corresponding to the same virtual resource instance are retrieved item by item, ensuring a continuous correspondence between the access request characteristics and the SSD channel occupancy information, media unit load information, and... Garbage collection activity information, wear and tear information, and write amplification change information are linked together to form a unified correspondence. Subsequently, each item with the established correspondence is organized item by item according to the virtual resource instance, so that the IO queue occupancy information and access request characteristics are respectively accompanied by the corresponding SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information before entering subsequent processing. After all the organization is completed, the IO queue occupancy information, access request characteristics, SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information that have been associated under the same virtual resource instance are written together to form the media bearing association result.

[0034] After the media bearer association results are formed, a grey relational analysis algorithm based on dynamic time warping alignment is performed on the load expression content corresponding to each virtual resource instance and the load expression content in the media operation status data. When receiving the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance, the IO queue occupancy information in the media bearer association results is read item by item according to the virtual resource instance. Then, the access request characteristics are read item by item according to the corresponding position consistent with the virtual resource instance, ensuring that the IO queue occupancy information and access request characteristics maintain a continuous correspondence under the same virtual resource instance. After the IO queue occupancy information and access request characteristics have been read, the IO queue occupancy information under the same virtual resource instance is arranged continuously according to the original corresponding order. Then, the access request characteristics under the same virtual resource instance are arranged continuously according to the corresponding order consistent with the IO queue occupancy information, ensuring that the access request characteristics always maintain a one-to-one correspondence with the IO queue occupancy information during the arrangement process. After the arrangement is completed, the IO queue occupancy information and access request characteristics are jointly organized and written to form a virtual resource load sequence. Virtual resource load value: The corresponding virtual resource load sequence formula is: ; : No. A virtual resource instance at time The virtual resource load value, : No. A virtual resource instance at time IO queue occupancy information : No. A virtual resource instance at time Access request characteristics, : No. The virtual resource load sequence corresponding to each virtual resource instance Timing length, Virtual resource instance number, : Timing position number.

[0035] Upon receiving SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information after they have been associated with each virtual resource instance, the system first reads the SSD channel occupancy information item by item according to the virtual resource instance. Then, it reads the media unit load information item by item according to the corresponding position of the same virtual resource instance, maintaining a continuous correspondence between the SSD channel occupancy information and the media unit load information. Subsequently, it continues to read the garbage collection activity information, wear and tear information, and write amplification change information item by item according to the same corresponding position, maintaining a correspondence between the garbage collection activity information, wear and tear information, and write amplification change information and the SSD channel occupancy information and media unit load information under the same virtual resource instance. After all information has been read, the SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information are arranged continuously in the original corresponding order, maintaining a one-to-one correspondence among the five types of information during the arrangement process. After the arrangement is completed, the SSD channel occupancy information, media unit load information, garbage collection activity information, wear and tear information, and write amplification change information are jointly organized and written to form a media operating load sequence. Media operating load value: The corresponding media operating load sequence: ; : No. A virtual resource instance at time The operating load value of the medium. : No. A virtual resource instance at time Corresponding SSD channel usage information : No. A virtual resource instance at time Corresponding media unit load information, : No. A virtual resource instance at time Information on corresponding waste recycling activities, : No. A virtual resource instance at time Corresponding wear and tear information, : No. A virtual resource instance at time The corresponding write amplification change information, : No. The media runtime load sequence corresponding to each virtual resource instance. Timing length, Virtual resource instance number, : Timing position number.

[0036] After the virtual resource load sequence and the media operation load sequence are formed, dimensionless processing is performed on both sequences. First, each load change in the virtual resource load sequence is read item by item, maintaining its original order during dimensionless processing. Then, each load change in the media operation load sequence is read item by item, maintaining its corresponding order with the virtual resource load sequence during dimensionless processing. After both sequences are read, a unified scale transformation is performed on each load change in the virtual resource load sequence, and the same transformation is performed on each load change in the media operation load sequence, converting them to a unified comparison scale. After the unified comparison scale transformation, the original correspondence between the virtual resource load sequence and the media operation load sequence remains unchanged. When performing dimensionless processing on each load change in the virtual resource load sequence, the range normalization formula is used: When performing dimensionless processing on each load change item in the media operating load sequence, the range normalization formula is used: ; : No. A virtual resource instance at time Dimensionless virtual resource load value, : No. A virtual resource instance at time Dimensionless media operating load value, : No. A virtual resource instance at time The virtual resource load value, : No. A virtual resource instance at time The operating load value of the medium. : No. Each virtual resource instance corresponds to the minimum value in the virtual resource load sequence. : No. Each virtual resource instance corresponds to the maximum value in the virtual resource load sequence. : No. Each virtual resource instance corresponds to the minimum value in the media runtime load sequence. : No. Each virtual resource instance corresponds to the maximum value in the media runtime load sequence. Virtual resource instance number, : Timing position number.

[0037] After dimensionless processing, dynamic time warping and alignment are performed on the virtual resource load sequence and the media operation load sequence. First, the load changes in the dimensionless virtual resource load sequence are read item by item according to each virtual resource instance. Then, the load changes in the dimensionless media operation load sequence are read item by item according to the corresponding position of the same virtual resource instance. After both types of load changes have been read, the load changes in the virtual resource load sequence are compared item by item with those in the media operation load sequence, and the process is continuously advanced along the corresponding arrangement direction of the virtual resource load sequence and the media operation load sequence. During this continuous advancement, the corresponding paths between the two sequences are adjusted item by item to ensure that the virtual resource load sequence and the media operation load sequence maintain temporal correspondence along the path with the minimum cumulative difference. After the path adjustment is completed, the corresponding results of the virtual resource load sequence and the media operation load sequence under the path with the minimum cumulative difference are organized and written to form the time alignment result. Dynamic time warping and alignment: Local distance: ;Redefine the cumulative distance recursive formula: The path with the minimum cumulative difference can be represented as: ; : No. The position of a virtual resource instance in the virtual resource load sequence With media operating load sequence position Local distance between them : No. A virtual resource instance at location The cumulative distance at the location, : No. A candidate alignment path for a virtual resource instance. : No. The path with the minimum cumulative difference for each virtual resource instance. : No. A virtual resource instance at location Dimensionless virtual resource load value, : No. A virtual resource instance at location Dimensionless media operating load value, Virtual resource load sequence position number, : Medium operating load sequence position number, Virtual resource instance number.

[0038] After the timing alignment results are formed, a difference sequence construction process is performed on the virtual resource load sequence and the media operation load sequence. First, the aligned virtual resource load changes are read item by item according to the timing alignment results. Then, the aligned media operation load changes are read item by item according to the corresponding positions of the virtual resource load changes. After both types of load changes have been read item by item, the aligned virtual resource load changes and media operation load changes are arranged in a corresponding order. Subsequently, difference extraction is performed on each aligned virtual resource load change and media operation load change, and the differences are written continuously in the original order of the timing alignment results. After all differences have been written, a difference sequence is formed. Difference sequence construction process: The corresponding differential sequence is: ; : No. The virtual resource instance is in the optimal alignment path. The difference value at each position, : No. The difference sequence corresponding to each virtual resource instance. : No. The virtual resource instance is in the optimal alignment path. Dimensionless virtual resource load value at each virtual resource location : No. The virtual resource instance is in the optimal alignment path. Dimensionless media operating load values ​​at each media location Optimal alignment path A virtual resource location, Optimal alignment path One medium location, : No. The optimal alignment path length for a virtual resource instance. : Optimal alignment path position number Virtual resource instance number.

[0039] After the difference sequence is formed, grey relational coefficient calculation is performed on it. First, each difference in the difference sequence is read item by item according to the virtual resource instance, maintaining its original order. Then, the correlation coefficient is calculated sequentially for each difference, resulting in a grey relational coefficient for each difference. After all differences are calculated, the grey relational coefficients are continuously organized and written back in their original order from the difference sequence, maintaining the correspondence between each grey relational coefficient and the corresponding virtual resource load change and media operation load change. Grey relational coefficient calculation process: ; : No. The virtual resource instance is in the optimal alignment path. Grey relational coefficients at each position : No. The virtual resource instance is in the optimal alignment path. The difference value at each position, The minimum value among all differences. The maximum value among all differences. Resolution factor : Optimal alignment path position number Virtual resource instance number.

[0040] After the grey relational coefficients are formed, grey relational degree calculation is performed based on them. First, the grey relational coefficients for each virtual resource instance are read item by item, and the grey relational coefficients under the same virtual resource instance are arranged consecutively. Then, the relational degree is calculated based on the consecutively arranged grey relational coefficients under the same virtual resource instance, and a set of grey relational coefficients corresponding to the same virtual resource instance forms a corresponding grey relational degree. After all grey relational degrees are formed, the grey relational degree corresponding to each virtual resource instance is written to each medium's operating status according to the original correspondence, forming the relational degree correspondence result. Grey relational degree calculation process: ; : No. A gray relational value corresponding to each virtual resource instance : No. The optimal alignment path length for a virtual resource instance. : No. The virtual resource instance is in the optimal alignment path. Grey relational coefficients at each position : Optimal alignment path position number Virtual resource instance number.

[0041] After the correlation degree correspondence results are formed, load coupling processing is performed on them. First, the gray correlation degrees in the correlation degree correspondence results are read item by item for each virtual resource instance. Then, the corresponding media operating states are read item by item according to the corresponding positions of the gray correlation degrees. After the gray correlation degrees and media operating states are read, the gray correlation degrees and media operating states under the same virtual resource instance are arranged continuously, ensuring that the original correspondence between the gray correlation degrees and media operating states is maintained. Subsequently, coupling processing is performed on the gray correlation degrees formed under different media operating states for the same virtual resource instance, ensuring that each gray correlation degree continuously corresponds to its corresponding media operating state during the processing. After coupling processing is completed, the processed gray correlation degrees and media operating states under the same virtual resource instance are written together to form the load coupling processing result. Load coupling processing: ; : No. The virtual resource instance and the first The results of load coupling sorting between the operating states of the media : No. The virtual resource instance and the first Grey correlation degree between the operating states of each medium : No. The number of media running states corresponding to each virtual resource instance. : Medium operating status number, The virtual resource instance number, from the perspective of the processing correspondence, corresponds to "forming a virtual resource load sequence, forming a media operation load sequence, dimensionless processing, dynamic time regularization and alignment processing, difference sequence construction processing, grey relational coefficient calculation processing, grey relational degree calculation processing, and load coupling sorting processing", which is consistent with the processing names in the implementation method.

[0042] After the load coupling adjustment results are generated, conflict relationship marking is performed on the resource competition relationships, media pressure overlap relationships, and load concentration relationships between virtual resource instances, between media operating states, and between virtual resource instances and media operating states. First, the correlation degree correspondence results in the load coupling adjustment results are read item by item for each virtual resource instance, and the correlation degree correspondence results under the same virtual resource instance are arranged continuously according to the correspondence relationship. Then, the correlation degree correspondence results between different virtual resource instances are compared and adjusted to form a resource competition relationship in the occupancy correlation content of multiple virtual resource instances under the same media operating state. After the resource competition relationship is formed, the SSD channel occupancy information and media unit load information under the same media operating state are processed. The coupling results corresponding to waste recycling activity information, wear change information, and write amplification change information are parallelized and organized to form a media pressure overlap relationship among multiple media pressure contents under the same media operating state. Subsequently, the coupling results of the same virtual resource instance in multiple media operating states are centrally organized to form a load concentration relationship among the associated centralized contents of the same virtual resource instance in multiple media operating states. After the resource competition relationship, media pressure overlap relationship, and load concentration relationship are formed respectively, the resource competition relationship, media pressure overlap relationship, and load concentration relationship are marked and written one by one, and each relationship maintains its original corresponding position between the corresponding virtual resource instance and media operating state, forming a conflict relationship marking result.

[0043] After the conflict relationship marking results are formed, the load coupling sorting results and conflict relationship marking results are integrated. First, the corresponding coupling content in the load coupling sorting results is read item by item according to the virtual resource instance. Then, the marking content in the conflict relationship marking results is read item by item according to the corresponding position consistent with the virtual resource instance, and the marking content and the corresponding coupling content are kept in the same virtual resource instance and the same medium operating state. Then, the correlation result, resource competition relationship, medium pressure overlap relationship, and load concentration relationship in the load coupling sorting results are merged and written with the corresponding marking content in the conflict relationship marking results. After all the content is merged, the integrated content of the same virtual resource instance and the same medium operating state is uniformly sorted and continuously output according to the virtual resource instance to generate bi-state mapping data.

[0044] In this embodiment, when the scheduling budget generation module receives the load coupling sorting results and conflict relationship marking results from the bi-state mapping data, it first reads the load coupling sorting results item by item according to the virtual resource instance, and then reads the conflict relationship marking results item by item according to the corresponding position consistent with the virtual resource instance, ensuring that the load coupling sorting results and conflict relationship marking results maintain a continuous correspondence under the same virtual resource instance. After the load coupling sorting results and conflict relationship marking results are read item by item, the load coupling sorting results under the same virtual resource instance are arranged continuously according to the original corresponding order, and then the conflict relationship marking results under the same virtual resource instance are arranged continuously according to the corresponding order consistent with the load coupling sorting results, ensuring that the conflict relationship marking results always maintain a sequential correspondence with the load coupling sorting results during the arrangement process. After the arrangement is completed, the load coupling sorting results and conflict relationship marking results are sorted according to the virtual resource instance, and the sorted content is continuously written to form the basic results of budget generation.

[0045] After the basic budget generation results are formed, business level association processing is performed on each virtual resource instance. First, the load coupling sorting results and conflict relationship marking results in the basic budget generation results are read item by item for each virtual resource instance. Then, the business level constraints are read item by item according to the corresponding position consistent with the virtual resource instance, and the business level constraints are kept in a continuous correspondence with the load coupling sorting results and conflict relationship marking results under the same virtual resource instance. After the business level constraints have been read, the business level constraints under the same virtual resource instance are arranged continuously according to the original corresponding order, and the business level constraints are kept in a continuous correspondence with the load coupling sorting results and conflict relationship marking results during the arrangement process. After the arrangement is completed, the business level constraints under the same virtual resource instance are associated with the basic budget generation results and written, so that each virtual resource instance and its corresponding business level constraints are associated. The content after the correspondence is established is written continuously to generate the business level association results.

[0046] Before and after the business level association results are generated, resource occupancy assessment processing is performed on the load coupling sorting results and conflict relationship marking results in the budget generation base results. First, the load coupling sorting results in the budget generation base results are read item by item according to virtual resource instances, ensuring a continuous reading relationship for load coupling sorting results under the same virtual resource instance. Then, the conflict relationship marking results are read item by item according to the corresponding positions consistent with the virtual resource instance, ensuring a one-to-one correspondence between the conflict relationship marking results and the load coupling sorting results. After the load coupling sorting results and conflict relationship marking results are read accordingly, the coupling corresponding content in the load coupling sorting results and the marking content in the conflict relationship marking results are compared and sorted, ensuring a continuous expansion of the corresponding content under the same virtual resource instance. During the continuous expansion process, ... The resource content under the same virtual resource instance is processed to adjust the resource contention level, and the resource contention level is continuously mapped to the virtual resource instance. Then, the media pressure level is processed to adjust the media pressure level to maintain a continuous mapping between the resource contention level and the media contention level under the same virtual resource instance. Next, the load concentration level is processed to adjust the load concentration level to maintain a continuous mapping between the resource contention level and the media pressure level under the same virtual resource instance. After all the processing is completed, the resource contention level, media pressure level, and load concentration level are continuously written to the corresponding virtual resource instance, so that the resource contention level, media pressure level, and load concentration level of each virtual resource instance form an evaluation correspondence, and the resource occupancy evaluation result is generated.

[0047] After the business level association results and resource occupancy assessment results are formed, the Jaya algorithm based on the simulated annealing acceptance mechanism is executed. When receiving the business level association results and resource occupancy assessment results, the business level constraints in the business level association results are read item by item according to the virtual resource instance. Then, the resource contention level, media pressure level, and load concentration level in the resource occupancy assessment results are read item by item according to the corresponding position consistent with the virtual resource instance, ensuring that the business level constraints and the resource contention level, media pressure level, and load concentration level maintain a continuous correspondence under the same virtual resource instance. After the business level constraints, resource contention level, media pressure level, and load concentration level are read item by item, the business level constraints are arranged continuously according to the original arrangement order of the virtual resource instance. Then, the resource contention level, media pressure level, and load concentration level are arranged continuously according to the corresponding order consistent with the business level constraints, ensuring that the resource contention level, media pressure level, and load concentration level always maintain a one-to-one correspondence with the business level constraints during the arrangement process. After the arrangement is completed, the business level association results and resource occupancy assessment results are combined and organized, and the combined and organized content is continuously written to the virtual resource instance to generate a candidate coordination solution set. Candidate Coordination Solution Set Generation Process: ; ; : The index of the corresponding position of the data in the sequence of virtual resource instances. : No. Business level constraints corresponding to each virtual resource instance : No. The degree of resource contention corresponding to each virtual resource instance. : No. The media pressure level corresponding to each virtual resource instance : No. The degree of load concentration corresponding to each virtual resource instance : No. The initial candidate coordination solution set for each virtual resource instance. : No. The first virtual resource instance One initial candidate coordination solution. The combination and sorting operator is used to write the business level constraints, resource contention level, media pressure level, and load concentration level into the first virtual resource instance in the corresponding order. One candidate coordinated solution The number of candidate coordination solutions under the same virtual resource instance.

[0048] After the candidate coordination solution set is formed, each candidate coordination solution in the set is read item by item according to the virtual resource instance, ensuring that candidate coordination solutions under the same virtual resource instance are read continuously. After all candidate coordination solutions under the same virtual resource instance have been read, the correspondence between each candidate coordination solution is organized item by item, and the candidate coordination solutions are continuously compared according to their order of arrangement under the same virtual resource instance. During the continuous comparison process, the merits of each candidate coordination solution under the same virtual resource instance are identified, and the identified best candidate coordination solution is written into the current best candidate coordination solution position according to its original position. After the current best candidate coordination solution is identified, the merits of each candidate coordination solution under the same virtual resource instance are further identified, and the identified worst candidate coordination solution is written into the current worst candidate coordination solution position according to its original position. After both the current best and the current worst candidate coordination solutions are identified, the current best and the current worst candidate coordination solutions are made to maintain a continuous correspondence with the remaining candidate coordination solutions under the same virtual resource instance. Current best candidate coordination solution identification: ; ; : No. During the round of iteration, the first The candidate coordination solution number with the best fitness under each virtual resource instance. : No. During the round of iteration, the first The current optimal candidate coordination solution for each virtual resource instance. : No. During the round of iteration, the first The first virtual resource instance Fitness values ​​of each candidate coordination solution. Identification of the worst candidate coordination solution: ; ; : No. During the round of iteration, the first The candidate coordination solution number with the worst fitness among the virtual resource instances. : No. During the round of iteration, the first The current worst candidate coordination solution for each virtual resource instance. : No. During the round of iteration, the first The first virtual resource instance The fitness values ​​of each candidate coordinated solution.

[0049] After identifying the current best and worst candidate coordination solutions, Jaya update processing is performed on the remaining candidate coordination solutions based on these two solutions. First, the current best candidate coordination solution is read item by item according to the virtual resource instance. Then, the current worst candidate coordination solution is read item by item according to the corresponding position within the same virtual resource instance, ensuring a continuous correspondence between the current best and worst candidate coordination solutions within the same virtual resource instance. After the current best and worst candidate coordination solutions are read, the remaining candidate coordination solutions are read item by item according to the corresponding position within the same virtual resource instance, ensuring a continuous correspondence between the remaining candidate coordination solutions and the current best and worst candidate coordination solutions within the same virtual resource instance. After establishing this correspondence, Jaya update processing is performed on the remaining candidate coordination solutions item by item, ensuring that each remaining candidate coordination solution is affected by both the current best and worst candidate coordination solutions during the update process. After all remaining candidate coordination solutions have been updated, the updated content is continuously organized and written according to the virtual resource instance, generating an updated candidate coordination solution set. Jaya Update Process: ; ; : No. During the round of iteration, the first The first virtual resource instance The first candidate coordination solution One portion, : No. During the round of iteration, the first The current optimal candidate coordination solution under the virtual resource instance is the One portion, : No. During the round of iteration, the first The worst candidate coordination solution under the given virtual resource instances is the... One portion, : No. During rounds of iteration, the optimal guiding random coefficients take values ​​ranging from 1 to 2. , : No. During rounds of iteration, the worst-case exclusion random coefficient takes values ​​ranging from 1 to 2. , : Component dimension of candidate coordinated solutions : No. The updated candidate coordinated solution during round iteration.

[0050] After the updated candidate coordination solution set is formed, a fitness comparison is performed between each updated candidate coordination solution and its corresponding original candidate coordination solution, and the fitness difference between the updated and original candidate coordination solutions is calculated. First, each updated candidate coordination solution in the updated candidate coordination solution set is read item by item according to the virtual resource instance. Then, the corresponding original candidate coordination solution in the candidate coordination solution set is read item by item according to the corresponding position consistent with the virtual resource instance, ensuring that each updated candidate coordination solution and its corresponding original candidate coordination solution maintain a continuous correspondence under the same virtual resource instance. In each updated candidate coordination solution... After reading the corresponding original candidate coordination solutions, a fitness comparison is performed on each updated candidate coordination solution and its corresponding original candidate coordination solution, forming a fitness comparison relationship between each updated candidate coordination solution and its corresponding original candidate coordination solution. After the fitness comparison is completed, difference calculation is performed on each fitness comparison relationship, forming a corresponding fitness difference between each updated candidate coordination solution and its corresponding original candidate coordination solution. After all fitness differences are formed, each updated candidate coordination solution, its corresponding original candidate coordination solution, and the fitness difference are continuously organized according to virtual resource instances. Fitness function: ; The fitness mapping function is used to map business level constraints, resource contention level, media pressure level, load concentration level, and the current candidate coordination solution to a single fitness value. Business level constraints The degree of competition for resources : Medium pressure level : Load concentration level : No. Candidate coordinated solutions during rounds of iteration. Fitness comparison processing: ; : No. During the round of iteration, the first The first virtual resource instance The comparison results between the updated candidate coordination solutions and the original candidate coordination solutions. Update the fitness values ​​of candidate coordinated solutions. : The fitness value of the original candidate coordinated solution, when When, it indicates that the updated candidate coordination solution is better than the original candidate coordination solution. When, it means that the updated candidate coordination solution is equal to the original candidate coordination solution in terms of fitness. When the updated candidate coordination solution is worse than the original candidate coordination solution, it indicates that the updated candidate coordination solution is inferior to the original candidate coordination solution.

[0051] After fitness comparison and fitness difference calculation are completed, when an updated candidate coordination solution is superior to its corresponding original candidate coordination solution, the updated candidate coordination solution retention process is performed. First, the superiority relationships in the fitness comparison process are identified item by item for each virtual resource instance. Then, updated candidate coordination solutions that form a superior relationship are retrieved item by item, ensuring that the updated candidate coordination solutions that form a superior relationship maintain a continuous correspondence with their corresponding original candidate coordination solutions within the same virtual resource instance. After the superiority relationship is identified, the updated candidate coordination solutions that form a superior relationship are written to the retention positions according to their original correspondences, ensuring that the updated candidate coordination solutions that form a superior relationship continue to maintain a continuous correspondence with their corresponding virtual resource instances after being written. After all updated candidate coordination solutions that form a superior relationship have been written, the updated candidate coordination solutions are retained. Fitness difference calculation process: ; : No. During the round of iteration, the first The first virtual resource instance The fitness difference between the updated candidate coordinated solution and the original candidate coordinated solution. Update the fitness values ​​of candidate coordinated solutions. : The fitness value of the original candidate coordinated solution, when When, it indicates that the updated candidate coordination solution is better than the original candidate coordination solution. When the updated candidate coordination solution is worse than the original candidate coordination solution, it indicates that the updated candidate coordination solution is inferior to the original candidate coordination solution.

[0052] When updating a candidate coordinated solution that is inferior to the corresponding original candidate coordinated solution, first, perform temperature decay update processing based on the current iteration round to form the current temperature; first, read the current iteration position item by item according to the current iteration round, then continuously perform temperature decay updates according to the advancement order of the current iteration round, ensuring that the temperature decay update results maintain a continuous correspondence with the current iteration round; after the temperature decay update is completed, write the temperature decay update results back to the current iteration round to form the current temperature. Candidate coordinated solution retention processing: ; : No. The round of iterations leads to a candidate coordinated solution with a reserved position. Update candidate coordinated solutions. Original candidate coordinated solution, : Results of fitness comparison.

[0053] After the current temperature is established, when an updated candidate coordination solution is inferior to its corresponding original candidate coordination solution, simulated annealing is performed based on the current temperature and fitness difference to determine acceptance. First, updated candidate coordination solutions that form an inferior relationship are read item by item according to the virtual resource instance. Then, the corresponding original candidate coordination solutions and their corresponding fitness differences are read item by item according to the corresponding positions consistent with the virtual resource instance, ensuring that the updated candidate coordination solutions, their corresponding original candidate solutions, and fitness differences that form an inferior relationship maintain a continuous correspondence within the same virtual resource instance. After the above correspondence is established, the current temperature is read item by item according to the corresponding positions consistent with the virtual resource instance, ensuring that the current temperature, the updated candidate coordination solutions that form an inferior relationship, their corresponding original candidate solutions, and fitness differences maintain a continuous correspondence within the same virtual resource instance. When the current temperature and fitness difference are... After the degree difference is read, simulated annealing is performed on each candidate coordination solution that forms a suboptimal relationship, and each candidate coordination solution is judged to either meet or not meet the acceptance conditions. When a candidate coordination solution meets the acceptance conditions, it is retained and continuously written to the retained position according to its original position. When a candidate coordination solution does not meet the acceptance conditions, it is discarded and removed from the subsequent retained sequence according to its original position. After all judgments are completed, the candidate coordination solutions that meet and do not meet the acceptance conditions retain their respective processing results. Temperature decay update processing: ; ; : No. The current temperature corresponding to the round of iterations. : No. The current temperature corresponding to the round of iterations. Temperature decay coefficient, with a value range of [value missing]. , Initial temperature. Simulated annealing acceptance criteria: ; ; : No. During the round of iteration, the first The first virtual resource instance The acceptance probability of a deteriorated update candidate coordinated solution. Update the fitness difference between the candidate coordinated solution and the original candidate coordinated solution. : No. The current temperature of the round of iterations, : Accept the judgment result. A value of 1 indicates that the acceptance condition is met, and a value of 0 indicates that the acceptance condition is not met. : No. The random number used for judgment during round iteration has a range of values. Update candidate coordinated solutions discarding process: ; : No. After rounds of iteration, the final output is the candidate coordinated solution from the set of retained candidate coordinated solutions. : Fitness comparison results The simulated annealing process will be used to determine the outcome. Update candidate coordinated solutions. Original candidate coordinated solution.

[0054] After completing the processes of retaining and discarding updated candidate reconciliation solutions, the process of identifying the current best candidate reconciliation solution, identifying the current worst candidate reconciliation solution, Jaya update processing, fitness comparison processing, and simulated annealing acceptance judgment is repeated based on the retained candidate reconciliation solution set until the iteration termination condition is met. First, all retained candidate reconciliation solutions are reorganized sequentially according to virtual resource instances, ensuring a stable corresponding order before entering the next round of processing. After the retained candidate reconciliation set is completed, the processes of identifying the current best candidate reconciliation solution, identifying the current worst candidate reconciliation solution, Jaya update processing, fitness comparison processing, and simulated annealing acceptance judgment are repeated on the retained candidate reconciliation solution set. After each round of repetition, the newly formed retained candidate reconciliation solution set continues as input for the next round of processing, maintaining a continuous iterative succession relationship between rounds. The current iteration round is continuously advanced, and the iteration termination condition is continuously checked during the advancement of the current iteration round. Once the iteration termination condition is met, the repetition stops. (Iteration Termination Condition Check)

[0055] : No. The termination condition for each iteration is determined by a value of 1, which indicates that the termination condition is met, and a value of 0 indicates that iteration continues. : Current iteration round Maximum number of iterations : No. During the first iteration The fitness value of the current best candidate coordinated solution under each virtual resource instance. : No. During the first iteration The fitness value of the current best candidate coordinated solution under each virtual resource instance. Termination tolerance is used to measure whether the change in optimal fitness between two adjacent rounds has become sufficiently small. This condition applies to all virtual resource instances simultaneously.

[0056] After the iteration termination condition is met, the current optimal candidate coordination solution retained at the time of meeting the iteration termination condition is determined as the media pressure coordination result. First, the current optimal candidate coordination solution retained at the time of meeting the iteration termination condition is read item by item according to the virtual resource instance. Then, the current optimal candidate coordination solution and its corresponding virtual resource instance are kept in a continuous correspondence. After the current optimal candidate coordination solution is read, the current optimal candidate coordination solution retained at the time of meeting the iteration termination condition is continuously written to the result position according to its original corresponding position, maintaining a continuous correspondence with the corresponding virtual resource instance during the writing process. After all writing is completed, the current optimal candidate coordination solution retained at the time of meeting the iteration termination condition is determined as the media pressure coordination result. Media pressure coordination result determination: ; : No. Media pressure coordination results corresponding to each virtual resource instance When the iteration termination condition is met, the 1st iteration... The current best candidate coordination solution retained under each virtual resource instance. The final iteration round that satisfies the iteration termination condition.

[0057] After the media pressure coordination results are generated, the scheduling allocation process is performed on the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth usage rhythm adjustment requirements corresponding to each virtual resource instance. When receiving the media pressure coordination results, each coordination content in the media pressure coordination results is read item by item according to the virtual resource instance, and the coordination content under the same virtual resource instance is kept in a continuous reading relationship. After the coordination content under the same virtual resource instance is read, the coordination content is arranged continuously according to its original corresponding order, and the coordination content under different virtual resource instances maintains its own independent corresponding position during the arrangement process. After the arrangement is completed, the media pressure coordination results are sorted according to the virtual resource instance, and the sorted content is written continuously according to the virtual resource instance to form the basic result of allocation generation.

[0058] After the basic results of the allocation generation are formed, namespace binding adjustment requirements are generated for the media pressure coordination results corresponding to each virtual resource instance. First, the media pressure coordination results in the basic results of the allocation generation are read item by item for each virtual resource instance, ensuring a continuous reading relationship for the media pressure coordination results under the same virtual resource instance. After the media pressure coordination results are read, the media pressure coordination results under the same virtual resource instance are continuously expanded in their original corresponding order, ensuring a continuous correspondence between the expanded items and the virtual resource instance. During the continuous expansion process, namespace binding changes are processed item by item for the media pressure coordination results under the same virtual resource instance, ensuring a one-to-one correspondence between each namespace binding change and its corresponding media pressure coordination result. After all namespace binding changes are processed, each namespace binding change is continuously written to the virtual resource instance, creating an adjustment correspondence between the namespace binding changes corresponding to each virtual resource instance, and generating namespace binding adjustment requirements.

[0059] After the namespace binding adjustment requirements are generated, virtual function allocation adjustment requirements are generated for the media pressure coordination results corresponding to each virtual resource instance. First, the media pressure coordination results in the basic allocation results are read item by item for each virtual resource instance, ensuring the read position matches the corresponding position of the virtual resource instance in the aforementioned namespace binding adjustment requirement generation process. After the media pressure coordination results are read, the media pressure coordination results under the same virtual resource instance are continuously expanded in their original corresponding order, ensuring that each expanded item maintains a continuous correspondence with the virtual resource instance. During the continuous expansion process, virtual function allocation changes are processed item by item for the media pressure coordination results under the same virtual resource instance, ensuring that each virtual function allocation change maintains a one-to-one correspondence with the corresponding media pressure coordination result. After all virtual function allocation changes are processed, each virtual function allocation change is continuously written to the virtual resource instance, creating an adjustment correspondence between the virtual function allocation changes corresponding to each virtual resource instance, and generating virtual function allocation adjustment requirements.

[0060] After the virtual function allocation adjustment requirements are generated, the queue delivery rhythm adjustment requirements are generated for the media pressure coordination results corresponding to each virtual resource instance. First, the media pressure coordination results in the basic results generated by the allocation ratio are read item by item according to the virtual resource instance, and the reading position is kept consistent with the corresponding position of the virtual resource instance in the aforementioned namespace binding adjustment requirements and virtual function allocation adjustment requirements. After the media pressure coordination results are read, the media pressure coordination results under the same virtual resource instance are continuously expanded in the original corresponding order, and the expanded items maintain a continuous correspondence with the virtual resource instance. During the continuous expansion process, the queue delivery change is sorted for each item of the media pressure coordination results under the same virtual resource instance, and each queue delivery change maintains a one-to-one correspondence with the corresponding media pressure coordination result. After all queue delivery changes are sorted, each queue delivery change is continuously written according to the virtual resource instance, so that the queue delivery changes corresponding to each virtual resource instance form an adjustment correspondence, and the queue delivery rhythm adjustment requirements are generated.

[0061] After the queue delivery rhythm adjustment requirement is generated, the bandwidth usage rhythm adjustment requirement generation process is performed on the media pressure coordination results corresponding to each virtual resource instance. First, the media pressure coordination results in the basic results of the allocation are read item by item according to the virtual resource instance, and the reading position is kept consistent with the corresponding position of the virtual resource instance in the aforementioned namespace binding adjustment requirement, virtual function allocation adjustment requirement, and queue delivery rhythm adjustment requirement. After the media pressure coordination results are read, the media pressure coordination results under the same virtual resource instance are continuously expanded in the original corresponding order, and the expanded items maintain a continuous correspondence with the virtual resource instance. During the continuous expansion process, the bandwidth usage change is sorted out item by item for the media pressure coordination results under the same virtual resource instance, and the bandwidth usage change is kept in a one-to-one correspondence with the corresponding media pressure coordination result. After all bandwidth usage changes are sorted out, the bandwidth usage changes are continuously written to the virtual resource instance, so that the bandwidth usage changes corresponding to each virtual resource instance form an adjustment correspondence, and the bandwidth usage rhythm adjustment requirement is generated.

[0062] After all namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth usage rhythm adjustment requirements are formed, corresponding integration processing is performed on these requirements. First, namespace binding adjustment requirements are read item by item for each virtual resource instance. Then, virtual function allocation adjustment requirements are read item by item according to the corresponding position consistent with that virtual resource instance, ensuring a continuous correspondence between namespace binding adjustment requirements and virtual function allocation adjustment requirements within the same virtual resource instance. Subsequently, the corresponding positions consistent with that virtual resource instance are used to... The system reads the queue delivery rhythm adjustment requirements and bandwidth usage rhythm adjustment requirements item by item, ensuring that these requirements maintain a continuous correspondence with the namespace binding adjustment requirements and virtual function allocation adjustment requirements within the same virtual resource instance. After all four adjustment requirements have been read, the system continuously integrates and writes these requirements by virtual resource instance, maintaining the original correspondence order within each virtual resource instance. Once all integration and writing are complete, the system generates scheduling budget data.

[0063] In this implementation, the local execution control module receives scheduling budget data and organizes the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth usage rhythm adjustment requirements in the scheduling budget data according to the corresponding virtual resource instances. First, it reads the namespace binding adjustment requirements corresponding to the same virtual resource instance item by item. Then, it reads the virtual function allocation adjustment requirements corresponding to the same virtual resource instance item by item, ensuring that the virtual function allocation adjustment requirements and namespace binding adjustment requirements maintain a correspondence within the same virtual resource instance. After the namespace binding adjustment requirements and virtual function allocation adjustment requirements are properly matched, the queue delivery rhythm adjustment requirements corresponding to the same virtual resource instance are executed. The rhythm adjustment requirements are read item by item, ensuring a continuous correspondence between queue delivery rhythm adjustment requirements, namespace binding adjustment requirements, and virtual function allocation adjustment requirements within the same virtual resource instance. Subsequently, bandwidth usage rhythm adjustment requirements corresponding to the same virtual resource instance are read item by item, ensuring a unified correspondence between bandwidth usage rhythm adjustment requirements, namespace binding adjustment requirements, virtual function allocation adjustment requirements, and queue delivery rhythm adjustment requirements. After all adjustment requirements have been read, the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth usage rhythm adjustment requirements are merged and organized by virtual resource instance to form the basic execution control results.

[0064] After the execution control foundation results are formed, namespace binding adjustment processing is performed on the namespace binding adjustment requirements corresponding to each virtual resource instance. First, the namespace binding adjustment requirements corresponding to the same virtual resource instance in the execution control foundation results are read, and the namespace binding adjustment requirements are kept in correspondence with the virtual resource instance when entering the adjustment process. Then, each binding change in the namespace binding adjustment requirements is retrieved item by item, and each binding change is continuously expanded according to the original corresponding order in the scheduling budget data. After continuous expansion, the namespace binding relationship is adjusted item by item for each binding change, and each namespace binding relationship adjustment result is kept in continuous correspondence with the corresponding virtual resource instance. After all namespace binding relationships are adjusted, the namespace binding relationship adjustment results under the same virtual resource instance are organized and written in the original order to form the namespace binding adjustment result.

[0065] After the namespace binding adjustment results are formed, virtual function allocation adjustment processing is performed on the virtual function allocation adjustment requirements corresponding to each virtual resource instance. First, the virtual function allocation adjustment requirements corresponding to the same virtual resource instance in the execution control basic results are read, and the correspondence between the virtual function allocation adjustment requirements and the virtual resource instance is maintained when the adjustment process begins. Then, each allocation change in the virtual function allocation adjustment requirements is retrieved item by item, and each allocation change is continuously expanded according to the original corresponding order in the scheduling budget data. After continuous expansion, the virtual function allocation relationship adjustment is performed on each allocation change item, and the virtual function allocation relationship adjustment result of each item is continuously corresponded to the corresponding virtual resource instance. After all virtual function allocation relationships are adjusted, the adjustment results of each virtual function allocation relationship under the same virtual resource instance are organized and written in the original order to form the virtual function allocation adjustment result.

[0066] After the virtual function allocation adjustment results are formed, queue delivery rhythm adjustment processing is performed on the queue delivery rhythm adjustment requirements corresponding to each virtual resource instance. First, the queue delivery rhythm adjustment requirements corresponding to the same virtual resource instance in the execution control basic results are read, and the correspondence between the queue delivery rhythm adjustment requirements and the virtual resource instance is maintained when the adjustment process is entered. Then, each delivery change content in the queue delivery rhythm adjustment requirements is retrieved item by item, and each delivery change content is continuously expanded according to the original corresponding order in the scheduling budget data. After continuous expansion, the queue delivery relationship adjustment is performed on each delivery change content item by item, and the queue delivery relationship adjustment result of each item is continuously corresponded with the corresponding virtual resource instance. After all queue delivery relationships are adjusted, the queue delivery relationship adjustment results under the same virtual resource instance are sorted and written in the original order to form the queue delivery rhythm adjustment result.

[0067] After the queue delivery rhythm adjustment results are formed, bandwidth usage rhythm adjustment processing is performed on the bandwidth usage rhythm adjustment requirements corresponding to each virtual resource instance. First, the bandwidth usage rhythm adjustment requirements corresponding to the same virtual resource instance in the execution control basic results are read, and the bandwidth usage rhythm adjustment requirements are kept in correspondence with the virtual resource instance when entering the adjustment process. Then, each change in bandwidth usage rhythm requirements is retrieved item by item, and each change in bandwidth usage rhythm requirements is continuously expanded according to the original corresponding order in the scheduling budget data. After continuous expansion, the bandwidth usage relationship is adjusted item by item for each change in bandwidth usage, and each bandwidth usage relationship adjustment result is kept in continuous correspondence with the corresponding virtual resource instance. After all bandwidth usage relationships are adjusted, the bandwidth usage relationship adjustment results under the same virtual resource instance are organized and written in the original order to form the bandwidth usage rhythm adjustment results.

[0068] After the namespace binding adjustment results, virtual function allocation adjustment results, queue delivery rhythm adjustment results, and bandwidth usage rhythm adjustment results are all formed, corresponding integration processing is performed on these results. First, the namespace binding adjustment results are read item by item according to the virtual resource instance. Then, the virtual function allocation adjustment results are read item by item according to the corresponding position consistent with the virtual resource instance, ensuring that the namespace binding adjustment results and the virtual function allocation adjustment results maintain a continuous correspondence under the same virtual resource instance. Subsequently, the queue delivery rhythm adjustment results and the bandwidth usage rhythm adjustment results are read item by item according to the corresponding position consistent with the virtual resource instance, ensuring that the queue delivery rhythm adjustment results and the bandwidth usage rhythm adjustment results maintain a continuous correspondence under the same virtual resource instance as the namespace binding adjustment results and the virtual function allocation adjustment results, respectively. After all adjustment results have been read, the namespace binding adjustment results, virtual function allocation adjustment results, queue delivery rhythm adjustment results, and bandwidth usage rhythm adjustment results are continuously integrated and written according to the virtual resource instance to generate a partial execution result.

[0069] The feedback update module receives partial execution results and organizes them according to the corresponding virtual resource instances. First, it reads the namespace binding adjustment results corresponding to the same virtual resource instance item by item. Then, it reads the virtual function allocation adjustment results corresponding to the same virtual resource instance item by item, ensuring a continuous correspondence between the virtual function allocation adjustment results and the namespace binding adjustment results under the same virtual resource instance. Next, it reads the queue delivery rhythm adjustment results and bandwidth usage rhythm adjustment results corresponding to the same virtual resource instance item by item, ensuring a continuous correspondence between the queue delivery rhythm adjustment results and the bandwidth usage rhythm adjustment results and the namespace binding adjustment results and the virtual function allocation adjustment results under the same virtual resource instance, respectively. After all results have been read, the namespace binding adjustment results, virtual function allocation adjustment results, queue delivery rhythm adjustment results, and bandwidth usage rhythm adjustment results are merged and organized according to the virtual resource instance to form the basic feedback update results.

[0070] After the feedback update base results are formed, the business completion delay aggregation processing is performed on the execution results corresponding to each virtual resource instance. First, the execution results corresponding to the same virtual resource instance in the feedback update base results are read, and the correspondence between each execution result and the virtual resource instance is maintained when entering the business completion delay aggregation processing. Then, the business completion delay content in each execution result is retrieved item by item, and the business completion delay content is continuously expanded according to the original corresponding order in the local execution results. After continuous expansion, the business completion delay content is continuously merged and written by virtual resource instance, and the business completion delay content under the same virtual resource instance maintains a continuous correspondence, forming the business completion delay aggregation result.

[0071] After the business completion latency aggregation results are formed, throughput change processing is performed on the execution results corresponding to each virtual resource instance. First, the execution results corresponding to the same virtual resource instance in the feedback update basic results are read, and the correspondence between each execution result and the virtual resource instance is maintained when entering the throughput change processing. Then, the throughput change content in each execution result is retrieved item by item, and the throughput change content is continuously expanded according to the original corresponding order in the local execution results. After continuous expansion, the throughput change content is continuously organized and written according to the virtual resource instance, and the throughput change content under the same virtual resource instance maintains a continuous correspondence, forming the throughput change processing result.

[0072] After the throughput change processing results are generated, a fair occupancy comparison is performed on the execution results corresponding to each virtual resource instance. First, the execution results corresponding to the same virtual resource instance in the feedback update base results are read, and the correspondence between each execution result and the virtual resource instance is maintained when entering the fair occupancy comparison process. Then, the resource occupancy content in each execution result is retrieved item by item, and the resource occupancy content is continuously expanded according to the original corresponding order in the local execution results. After continuous expansion, the resource occupancy content under the same virtual resource instance is compared item by item, and the continuous correspondence between the resource occupancy content is maintained during the comparison process. After all comparisons are completed, the resulting resource occupancy differences are continuously written to the virtual resource instances to form the fair occupancy comparison results.

[0073] After the fair occupancy comparison results are generated, media lifetime change correlation processing is performed on the execution results corresponding to each virtual resource instance. First, the execution results corresponding to the same virtual resource instance in the feedback update base results are read, and the correspondence between each execution result and the virtual resource instance is maintained when entering the media lifetime change correlation processing. Then, the media lifetime change content in each execution result is retrieved item by item, and the media lifetime change content is continuously expanded according to the original corresponding order in the local execution results. After continuous expansion, the media lifetime change content is correlated and organized item by item, and the media lifetime change content is kept in continuous correspondence with the corresponding virtual resource instance during the organization process. After all media lifetime change content has been correlated and organized, the resulting lifetime changes are continuously written according to the virtual resource instance to form the media lifetime change correlation result.

[0074] After all the results of service completion delay collection, throughput change processing, fair occupancy comparison, and media lifetime change correlation are generated, corresponding integration processing is performed on these results. First, the service completion delay collection results are read item by item for each virtual resource instance. Then, the throughput change processing results are read item by item according to the corresponding position consistent with that virtual resource instance, ensuring a continuous correspondence between the service completion delay collection results and the throughput change processing results within the same virtual resource instance. Finally, the corresponding positions consistent with that virtual resource instance are used to integrate the results. The system reads the fair occupancy comparison results and media lifetime change correlation results item by item, ensuring that these results maintain a continuous correspondence with the service completion delay collection results and throughput change processing results within the same virtual resource instance. After all results have been read, the system continuously integrates and writes the service completion delay collection results, throughput change processing results, fair occupancy comparison results, and media lifetime change correlation results by virtual resource instance to generate service assurance feedback data. After the service assurance feedback data is generated, it is continuously output to the scheduling budget generation module by virtual resource instance.

[0075] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An enterprise-level SSD resource scheduling system based on storage virtualization, characterized in that: include: The virtual resource modeling module performs aggregation and association processing, business level binding processing, and access feature sorting processing on tenant identifiers, business volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics to generate virtual resource status data. The media status sensing module performs corresponding disk-based sorting, time-sequence organization, and media pressure merging processing on SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information to generate media operating status data. The dual-state mapping construction module performs resource instance correspondence processing and media bearer association processing on virtual resource status data and media operation status data. It then performs gray relational analysis algorithm based on dynamic time warping alignment and conflict relationship marking processing on the content after resource instance correspondence processing and media bearer association processing to generate dual-state mapping data. The scheduling budget generation module performs business level association processing, resource occupancy assessment processing, Jaya algorithm processing based on simulated annealing acceptance mechanism processing, and scheduling ratio generation processing on the bi-state mapping data to generate scheduling budget data; The local execution control module performs namespace binding adjustment, virtual function allocation adjustment, queue delivery rhythm adjustment, and bandwidth usage rhythm adjustment on the scheduling budget data, and generates local execution results. The feedback update module performs business completion delay aggregation processing, throughput change sorting processing, fair occupancy comparison processing, and media lifetime change correlation processing on the local execution results, generates service assurance feedback data, and outputs the service assurance feedback data to the scheduling budget generation module.

2. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 1, characterized in that: The virtual resource modeling module receives tenant identifiers, service volume identifiers, namespace identifiers, virtual function identifiers, IO queue occupancy information, and access request characteristics. It then aggregates and organizes these identifiers to form resource identifier aggregation results. Based on the resource identifier aggregation results, business level binding is performed on the tenant identifier, business volume identifier, namespace identifier and virtual function identifier corresponding to each virtual resource instance, so that each virtual resource instance is associated with the corresponding business level constraint, thus forming a business level binding result. Based on the resource identifier aggregation results and business level binding results, access feature processing is performed on the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance, so that the request load, request arrival status and queue occupancy status corresponding to each virtual resource instance are correlated, and virtual resource status data is generated.

3. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 1, characterized in that: The media status awareness module receives SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information, and performs corresponding on-disk organization on the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information in chronological order to form media status organization results; Based on the media status sorting results, the SSD channel occupancy information and media unit load information corresponding to each SSD are processed in a time sequence to form a sequential correspondence between the changes in SSD channel occupancy and media unit load at different times, thus forming the occupancy and load organization results. Based on the media status sorting results and occupancy load organization results, media pressure merging processing is performed on garbage collection activity information, wear change information, and write amplification change information to establish a pressure correspondence between the garbage collection changes, wear changes, and write amplification changes corresponding to each SSD and the SSD channel occupancy changes and media unit load changes, thereby generating media operating status data.

4. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 1, characterized in that: The dual-state mapping construction module receives tenant identifier, service volume identifier, namespace identifier, virtual function identifier, IO queue occupancy information and access request characteristics from the virtual resource status data, and receives SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information and write amplification change information from the media operation status data. It performs resource instance mapping and organization on the virtual resource status data and the media operation status data to form resource instance mapping results. Based on the resource instance correspondence results, the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance are combined with the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information and write amplification change information in the resource instance correspondence results to perform bearer association and organize them to form media bearer association results. Based on the media bearing association results, the load expression content corresponding to each virtual resource instance and the load expression content in the media operation status data are processed by a grey relational analysis algorithm based on dynamic time warping alignment to form a load coupling sorting result; Based on the load coupling sorting results, conflict relationship marking processing is performed on the resource competition relationship, media pressure overlap relationship and load concentration relationship between virtual resource instances, between media operating states and between virtual resource instances and media operating states to form conflict relationship marking results; Based on the results of load coupling sorting and conflict relationship marking, corresponding integration processing is performed to generate bi-state mapping data.

5. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 4, characterized in that: The grey relational analysis algorithm based on dynamic time warping alignment processes the following: based on the media bearer association results, it receives the IO queue occupancy information and access request characteristics corresponding to each virtual resource instance, as well as the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information after completing the bearer association with each virtual resource instance. It then organizes the IO queue occupancy information and access request characteristics into a virtual resource load sequence, and organizes the SSD channel occupancy information, media unit load information, garbage collection activity information, wear change information, and write amplification change information into a media operation load sequence. Dimensionless processing is performed on the virtual resource load sequence and the media operation load sequence to convert the load changes in the virtual resource load sequence and the load changes in the media operation load sequence into a unified comparison scale. Dynamic time warping and alignment are performed on the virtual resource load sequence and media operation load sequence that have been dimensionless processed. The virtual resource load sequence and media operation load sequence are time-aligned according to the path with the minimum cumulative difference to form a time-aligned result. Based on the time-series alignment results, differential sequence construction processing is performed on the virtual resource load sequence and the media operation load sequence to make the virtual resource load change content that has completed time-series alignment correspond to the media operation load change content item by item, forming a differential sequence. The grey relational coefficient is calculated based on the difference sequence, and the grey relational degree is calculated based on the grey relational coefficient, so that a correlation degree correspondence is formed between each virtual resource instance and each media operation status. Based on the correlation degree corresponding results, load coupling sorting is performed to form the load coupling sorting result.

6. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 1, characterized in that: The scheduling budget generation module receives the load coupling sorting results and conflict relationship marking results from the dual-state mapping data, and performs corresponding sorting according to the virtual resource instances to form the basic results for budget generation; Based on the budget generation results, business level association processing is performed on each virtual resource instance to establish a correspondence between each virtual resource instance and its corresponding business level constraints, thereby generating business level association results. Based on the budget generation results, resource occupancy assessment is performed on the load coupling sorting results and conflict relationship marking results corresponding to each virtual resource instance, so that the resource contention degree, media pressure degree and load concentration degree corresponding to each virtual resource instance form an assessment correspondence, and the resource occupancy assessment results are generated. Based on the business level correlation results and resource occupancy assessment results, the Jaya algorithm based on the simulated annealing acceptance mechanism is executed to generate media pressure coordination results; Based on the media pressure coordination results, the scheduling ratio generation process is performed on the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth usage rhythm adjustment requirements corresponding to each virtual resource instance, and scheduling budget data is generated.

7. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 6, characterized in that: The Jaya algorithm processing based on the simulated annealing acceptance mechanism includes: receiving the service level association results and resource occupancy assessment results, and combining and organizing the service level association results and resource occupancy assessment results to generate a set of candidate coordination solutions; Based on the candidate coordination solution set, the current best candidate coordination solution and the current worst candidate coordination solution are identified, and the remaining candidate coordination solutions are updated using Jaya based on the current best candidate coordination solution and the current worst candidate coordination solution to generate an updated candidate coordination solution set; For each updated candidate coordination solution in the updated candidate coordination solution set, a fitness comparison is performed with the corresponding original candidate coordination solution, and the fitness difference between the updated candidate coordination solution and the original candidate coordination solution is calculated; when the updated candidate coordination solution is better than the corresponding original candidate coordination solution, the updated candidate coordination solution is retained; Temperature decay update processing is performed based on the current iteration round to form the current temperature; when the updated candidate coordination solution is inferior to the corresponding original candidate coordination solution, simulated annealing acceptance judgment is performed based on the current temperature and fitness difference. Updated candidate coordination solutions that meet the acceptance conditions are retained, and updated candidate coordination solutions that do not meet the acceptance conditions are discarded. Based on the retained set of candidate reconciliation solutions, the process of identifying the current best candidate reconciliation solution, identifying the current worst candidate reconciliation solution, Jaya update processing, fitness comparison processing, and simulated annealing acceptance judgment is repeated until the iteration termination condition is met. The current optimal candidate coordination solution that is retained when the iteration termination condition is met is determined as the medium pressure coordination result.

8. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 6, characterized in that: The scheduling allocation generation process includes: receiving the media pressure coordination results and performing corresponding processing according to the virtual resource instances to form the basic results for allocation generation; Based on the basic results of the ratio generation, namespace binding adjustment requirements are generated for the media pressure coordination results corresponding to each virtual resource instance, so that the namespace binding changes corresponding to each virtual resource instance form an adjustment correspondence. Based on the basic results of the ratio generation, the virtual function allocation adjustment requirements generation process is performed on the media pressure coordination results corresponding to each virtual resource instance, so that the virtual function allocation changes corresponding to each virtual resource instance form an adjustment correspondence. Based on the basic results of the ratio generation, the queue delivery rhythm adjustment requirement generation process is performed on the media pressure coordination results corresponding to each virtual resource instance, so that the queue delivery changes corresponding to each virtual resource instance form an adjustment correspondence. Based on the basic results of the allocation ratio, the bandwidth usage rhythm adjustment requirement generation process is performed on the media pressure coordination results corresponding to each virtual resource instance, so that the bandwidth usage changes corresponding to each virtual resource instance form an adjustment correspondence. Based on the namespace binding adjustment requirements, virtual function allocation adjustment requirements, queue delivery rhythm adjustment requirements, and bandwidth usage rhythm adjustment requirements, corresponding integrated processing is performed to generate scheduling budget data.

9. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 1, characterized in that: The local execution control module receives namespace binding adjustment requests, virtual function allocation adjustment requests, queue delivery rhythm adjustment requests, and bandwidth usage rhythm adjustment requests from the scheduling budget data, and organizes them according to the virtual resource instances to form the basic execution control results. Based on the execution control foundation results, namespace binding adjustment processing is performed on the namespace binding adjustment requirements corresponding to each virtual resource instance, so that the namespace binding relationship corresponding to each virtual resource instance forms the adjustment result; Based on the execution control results, virtual function allocation adjustment processing is performed on the virtual function allocation adjustment requirements corresponding to each virtual resource instance, so that the virtual function allocation relationship corresponding to each virtual resource instance forms the adjustment result. Based on the execution control results, the queue delivery rhythm adjustment process is performed on the queue delivery rhythm adjustment requirements corresponding to each virtual resource instance, so that the queue delivery relationship corresponding to each virtual resource instance forms an adjustment result. Based on the execution control results, bandwidth usage rhythm adjustment processing is performed on the bandwidth usage rhythm adjustment requirements corresponding to each virtual resource instance, so that the bandwidth usage relationship corresponding to each virtual resource instance forms the adjustment result. Based on the results of namespace binding adjustment, virtual function allocation adjustment, queue delivery rhythm adjustment, and bandwidth usage rhythm adjustment, corresponding integration processing is performed to generate local execution results.

10. The enterprise-level SSD resource scheduling system based on storage virtualization according to claim 1, characterized in that: The feedback update module receives the partial execution results and organizes them according to the virtual resource instance to form the basic feedback update results; Based on the feedback update of the basic results, the business completion delay collection processing is performed on the execution results corresponding to each virtual resource instance, so that the business completion delay corresponding to each virtual resource instance forms a collection result. Based on the feedback update of the basic results, the throughput change processing is performed on the execution results corresponding to each virtual resource instance to form a processing result of the throughput change corresponding to each virtual resource instance. Based on the feedback update of the basic results, fair occupancy comparison processing is performed on the execution results corresponding to each virtual resource instance, so that the resource occupancy differences corresponding to each virtual resource instance form a comparison result. Based on the feedback update of the basic results, media lifetime change correlation processing is performed on the execution results corresponding to each virtual resource instance, so that the lifetime changes corresponding to each virtual resource instance form a correlation result. Based on the results of business completion delay aggregation, throughput change processing, fair occupancy comparison, and media lifetime change correlation, corresponding integration processing is performed to generate service assurance feedback data, which is then output to the scheduling budget generation module.

Images