Cloud resource dynamic deployment method and system supporting multi-product intelligent linkage

Abstract

This invention discloses a method and system for dynamic deployment of cloud resources supporting intelligent linkage of multiple products. In the field of cloud computing technology, this invention collects billing window information for each instance, maps it to time interval segments with boundary offsets, and simultaneously marks overlapping phase difference areas to construct a cross-instance billing phase correlation graph. When instance expansion needs are detected, the optimal supply instance is selected based on this graph. During scheduling, cross-product computing power borrowing and deployment within the billing phase difference window is completed through memory hot migration and traffic remapping. This invention enables refined reuse of paid idle computing power, improves resource scheduling accuracy and utilization, and adapts to the deployment needs of cloud resources with intelligent linkage of multiple products.

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, specifically to a method and system for dynamic deployment of cloud resources that supports intelligent linkage among multiple products. Background Technology

[0002] With the continuous improvement of the cloud service ecosystem, cloud resource scheduling with multi-product collaboration has become a core requirement for enterprises to reduce costs and increase efficiency. Most existing dynamic cloud resource deployment technologies are built around resource load and business traffic indicators to create elastic scheduling systems, which can achieve basic on-demand resource allocation. However, in multi-product collaboration scenarios, the following shortcomings still exist: First, existing technologies ignore the uninterruptibility of billing windows and the sunk cost effect under the pay-as-you-go model of cloud resources. They cannot quantify the boundary offset and phase difference overlap area of ​​billing windows of different instances, making it difficult to achieve fine-grained reuse of paid idle computing power between cross-product instances, resulting in ineffective waste of computing power resources. Secondly, the existing cross-instance resource scheduling lacks global billing phase correlation topology support. The matching of supply and demand instances only relies on static resource specification verification and does not comprehensively select the best based on multiple dimensions such as effective availability time, migration feasibility, and billing sunk value, resulting in insufficient matching accuracy. Therefore, there is an urgent need for cloud resource dynamic deployment methods and systems that support intelligent linkage among multiple products. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for dynamic deployment of cloud resources that supports intelligent linkage among multiple products, solving the problems of low cloud resource reuse rate and high scheduling cost under multi-product linkage.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a dynamic deployment method for cloud resources supporting intelligent linkage of multiple products, comprising: Step 1: Collect the start timestamp and billing granularity unit of the current billing window of the cloud account where each product instance is located, map the billing window of each instance to a time interval line segment with boundary offset, and mark the phase difference overlap area at the end of the billing window of different instances to construct a cross-instance billing phase correlation map. Step 2: Monitor the instantaneous resource demand queue length of demand-side instance B. When the queue increase exceeds the preset expansion threshold, query all supply-side instances that have an associated edge with instance B from the cross-instance billing phase association graph, and determine whether the start and end timestamps of the overlapping area recorded in the associated edge of each supply-side instance cover the current time. If they cover, extract the supply-side instance A corresponding to the associated edge, calculate the remaining effective duration and the available idle computing power specification of instance A in the overlapping area, and match and verify it with the amount of temporary expansion resources required by instance B to select the most suitable supply-side instance A1. Step 3: Issue a memory hot migration command to the control plane of the host machine where the most suitable supplier instance A1 is located, and asynchronously copy the memory pages of A1 to the reserved container slots of the resource pool where B is located. At the same time, update the virtual network forwarding plane to remap some traffic addresses pointing to A1 to temporary slots of B, and complete the cross-product borrowing and deployment of computing power within the billing phase difference window.

[0005] As a further aspect of the present invention, the specific operation of mapping the billing window of each instance to a time interval line segment with a boundary offset is as follows: Obtain the billing granularity unit G and the start time T_start of the current billing window for the instance from the instance metadata service or metering API interface provided by the cloud vendor; With T_start as the reference origin, the billing window of this instance is represented as a closed interval lnterval=[T_start,T_start+G]; Calculate the difference between the actual start timestamp of the instance and the start time T_start, and use it as the boundary offset Offset; Construct a time interval line segment Line={lnterval,Offset} carrying boundary offsets.

[0006] As a further aspect of the present invention, the specific steps for identifying the phase difference overlap region appearing at the end of the billing window for different instances are as follows: From the established set of time interval segments, extract the billing window end sub-interval for each product instance one by one, and combine the end sub-intervals of all product instances into a set of intervals to be processed. The starting point of the terminal sub-interval is the time point corresponding to the preset duration threshold △T that is traced back from the end time of the billing window of this instance, and the ending point is the end time of the billing window. The preset duration threshold △T represents the maximum remaining duration for which the system determines a billing window has entered the end state; Let any one of the end sub-intervals in the set of operation intervals be denoted as the supply-side instance A, and another any one of the end sub-intervals be denoted as the demand-side instance B. When the intersection of the end sub-intervals of A and B on the time axis is >0, calculate the first coverage rate of the intersection duration as a percentage of the duration of the end sub-interval of the supply-side instance A, and the second coverage rate as a percentage of the duration of the end sub-interval of the demand-side instance B. If the first coverage rate is greater than or equal to the preset first threshold value, and the second coverage rate is greater than or equal to the preset second threshold value, then the intersection is marked as a phase difference overlap region, and the start and end timestamps, duration, and identifiers of the supplier instance A and demand instance B of the overlap region are recorded.

[0007] As a further aspect of the present invention, the specific operation for constructing a cross-instance billing phase correlation graph is as follows: All the marked phase difference overlap areas are used as edge sets, and the supply-side instance and demand-side instance recorded in each phase difference overlap area are used as nodes to construct a cross-instance billing phase association graph. Each edge of the cross-instance billing phase association graph is represented by a directed edge pointing from the supply-side instance node to the demand-side instance node corresponding to the overlapping area, and an edge attribute set is attached. The set of edge attributes includes: the start and end timestamps of the overlapping area, the duration of the overlap, the first coverage rate, and the idle computing power specifications that the supplier instance can release within the overlapping area. When the billing window of any product instance changes, a partial update of the cross-instance billing phase association graph is triggered.

[0008] As a further aspect of the present invention, the specific rules for triggering local updates of the cross-instance billing phase correlation graph are as follows: If the last sub-interval of the billing window corresponding to the product instance has slipped out of the instance at the current time, then remove all edges with that instance as the supply-side node or demand-side node. If a new phase difference overlap region is identified based on the end sub-interval of the billing window corresponding to the product instance, then a directed edge is added between the nodes of the product instance and the edge attribute set is filled.

[0009] As a further aspect of the present invention, the specific steps for calculating the remaining effective duration and the available idle computing power specification of instance A within the overlapping region are as follows: Extract the start and end timestamps of the overlapping area between supplier instance A and demand instance B from the cross-instance billing phase correlation graph, and use the difference between the end time T_end of the overlapping area and the current physical time T_current as the base remaining duration △t_base. The base remaining time △t_base is corrected using the boundary offset Offset of the supplier instance A, and the effective remaining time △t_valid=min(△t_base,GA-Offset-△t_elapsed) is calculated, where GA is the billing granularity unit of supplier instance A, and △t_elapsed is the total running time of supplier instance A since the start of the current billing window. Get the average CPU utilization U_cpu and average memory usage U_mem of supplier instance A within the preset sampling window before the current time, and get the specification configuration of the instance from the instance metadata, including the number of vCPU cores C_total and the total memory M_total; Calculate the number of freed idle vCPU cores C_idle=C_total×(1-U_cpu)×ɑ, and the freed idle memory capacity M_idle=M_total×(1-U_mem)×β, where ɑ and β are preset borrowable scaling factors; The available idle computing power specification is obtained as Spec={vCPU:C_idle, memory:M_idle}.

[0010] As a further aspect of the present invention, the specific steps for matching and verifying the amount of temporary expansion resources required by instance B are as follows: Taking the available idle computing power specification A_Spec and the effective remaining duration A_△t_valid of supplier instance A as input, and comparing them with the temporary expansion resource requirement B_Spec and expansion duration requirement B_△t_valid of demand instance B, a three-dimensional verification is performed: Dimension 1: Verify whether the resource specification is met, i.e., A_C_idle ≥ B_C_idle and A_M_idle ≥ B_M_idle; Dimension 2: Verify whether the duration is met, i.e., A_△t_valid ≥ B_△t_valid; Dimension 3: Verify whether the borrowing feasibility is met, i.e. Where △t_migrate is the execution time required for the borrowing operation. This is the preset tolerance coefficient for the secondment execution time; The supplier instance A that simultaneously meets the above three verification conditions is marked as a candidate supplier instance. For all marked candidate supplier instances, the comprehensive matching score of each candidate instance is calculated, and the supplier instance A with the highest comprehensive matching score is selected as the most suitable supplier instance A1.

[0011] As a further aspect of the present invention, the specific operation for calculating the comprehensive matching score of each candidate instance is as follows: Calculate the resource specification matching degree S_spec, duration coverage S_time, and billing sunk value S_cost for each candidate instance; The resource specification matching degree S_spec is taken as the smaller value of the ratio between the supply specification and the demand specification in the two dimensions of vCPU cores and memory capacity, that is, S_spec=min(A_C_idle / B_C_idle,A_M_idle / B_M_idle); The duration coverage rate S_time is the ratio of the effective remaining duration to the required duration, i.e., S_time = A_△t_valid / B_△t_valid; The sunk cost of the billing is the ratio of the duration of overlap to the billing granularity of the supply instance. Calculate the comprehensive matching score Grade for each candidate instance: Grade = q1 × S_spec + q2 × S_time + q3 × S_cost, where q1, q2, and q3 are preset weights that satisfy q1 + q2 + q3 = 1 and q1 > q2 > q3.

[0012] As a further aspect of the present invention, the specific rule for remapping a portion of the traffic address pointing to A1 to the temporary slot of B is as follows: Extract the set of target traffic addresses that conform to the preset traffic identification rules from the virtual network forwarding plane of A1; The traffic identification rules include: the target IP address is a virtual IP of A1 and the target port belongs to a preset set of available ports; the source IP address belongs to a preset tenant network segment; and the HTTP header carries a preset tenant identifier. The next-hop address in the forwarding routing table entry corresponding to the target traffic address set is updated to the network address of B's ​​temporary slot, so that the matching traffic that subsequently arrives at the virtual network forwarding plane is forwarded to B's temporary slot.

[0013] A cloud resource dynamic deployment system that supports intelligent linkage among multiple products, including: The graph construction module collects the start timestamp and billing granularity unit of the current billing window of the cloud account where each product instance is located, maps the billing window of each instance to a time interval line segment with boundary offset, and marks the phase difference overlap area at the end of the billing window of different instances through interval algebra operations, and constructs a cross-instance billing phase correlation graph. The instance filtering module monitors the instantaneous resource demand queue length of demand-side instance B. When the queue increase exceeds the preset expansion threshold, it queries all supply-side instances that have an associated edge with instance B from the cross-instance billing phase association graph. It then checks whether the start and end timestamps of the overlapping area recorded in the associated edge of each supply-side instance cover the current time. If they cover, it extracts the supply-side instance A corresponding to the associated edge, calculates the remaining effective duration and available idle computing power specification of instance A in the overlapping area, and matches and verifies them with the amount of temporary expansion resources required by instance B to filter out the most suitable supply-side instance A1. The temporary scheduling module issues a memory hot migration instruction to the control plane of the host machine where the most suitable supplier instance A1 is located, and asynchronously copies the memory pages of A1 to the reserved container slots in the resource pool where B is located. At the same time, it updates the virtual network forwarding plane to remap some traffic addresses pointing to A1 to the temporary slots of B, thus completing the cross-product borrowing and deployment of computing power within the billing phase difference window.

[0014] This invention provides a method and system for dynamic deployment of cloud resources that supports intelligent linkage among multiple products, and has the following advantages compared with the prior art: (1) By constructing a cross-instance billing phase correlation graph, this invention incorporates time-dimensional information such as billing window, boundary offset, and phase difference overlap area into the resource scheduling system, which can accurately identify paid idle computing power at the end of the billing window, realize refined reuse of computing power across product instances, and reduce the ineffective waste of computing power resources. (2) This invention uses a multi-dimensional verification and comprehensive matching degree scoring mechanism, and comprehensively considers core factors such as resource specifications, effective availability, migration feasibility and billing sunk value to screen the optimal supply instance and effectively avoid the risk of scheduling failure. (3) This invention achieves lossless cross-product computing power borrowing and deployment within the billing phase difference window through memory hot migration and traffic remapping mechanism, and can complete temporary computing power expansion without adding billing costs, thus meeting the refined deployment needs of intelligent linkage of multiple products. Attached Figure Description

[0015] Figure 1 This is a flowchart of the steps of the present invention; Figure 2 This is the system principle block diagram of the present invention. Detailed Implementation

[0016] 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.

[0017] like Figure 1 This invention provides a dynamic deployment method for cloud resources that supports intelligent linkage of multiple products; As an embodiment of this application, it includes: Step 1: Collect the start timestamp and billing granularity unit of the current billing window of the cloud account where each product instance is located, map the billing window of each instance to a time interval line segment with boundary offset, and use interval algebra operations to mark the phase difference overlap area at the end of the billing window of different instances, and construct a cross-instance billing phase correlation map. Step 2: Monitor the instantaneous resource demand queue length of demand-side instance B. When the queue increase exceeds the preset expansion threshold, query all supply-side instances that have an associated edge with instance B from the cross-instance billing phase association graph, and determine whether the start and end timestamps of the overlapping area recorded in the associated edge of each supply-side instance cover the current time. If they cover, extract the supply-side instance A corresponding to the associated edge, calculate the remaining effective duration and the available idle computing power specification of instance A in the overlapping area, and match and verify it with the amount of temporary expansion resources required by instance B to select the most suitable supply-side instance A1. Step 3: Issue a memory hot migration command to the control plane of the host machine where the most suitable supplier instance A1 is located, and asynchronously copy the memory pages of A1 to the reserved container slots of the resource pool where B is located. At the same time, update the virtual network forwarding plane to remap some traffic addresses pointing to A1 to temporary slots of B, and complete the cross-product borrowing and deployment of computing power within the billing phase difference window.

[0018] As a second embodiment of this application, it is implemented based on the first embodiment, except that this embodiment includes: Step 1: Collect the start timestamp and billing granularity unit of the current billing window for each product instance in the cloud account. In existing cloud resource scheduling technologies, decision-making is usually limited to runtime performance indicators such as CPU utilization, memory usage, and network bandwidth, as well as cost accounting results such as resource unit price and total budget. However, this scheduling model ignores the fact that the billing window for cloud resources on a pay-as-you-go basis is uninterrupted and has sunk cost effects. Specifically, for a cloud server billed by the hour, even if it is released in the last second of the billing window, the user still needs to pay for the entire hour. By collecting the start timestamp of the billing window, the system can record the remaining time of each product instance from the current moment to its next billing settlement node, thereby injecting time information that originally belonged to the financial level into the real-time resource deployment decision flow. The billing granularity unit is the smallest scale by which cloud providers quantify resource usage time. Common billing granularity units include: hours, which are typically used for long-running computing resources such as ECS and bare metal servers; seconds, which are widely used for container instances, function compute, and some cloud providers' pay-as-you-go cloud disks and databases; minutes, which are often seen in early cloud products or some specific GPU instances; in addition, there are granularities for daily or monthly billing of network traffic. The billing window for each instance is mapped to a time interval segment with a boundary offset, specifically as follows: Obtain the billing granularity unit G and the start time T_start of the current billing window for the instance from the instance metadata service or metering API interface provided by the cloud vendor; With T_start as the reference origin, the billing window of this instance is represented as a closed interval lnterval=[T_start,T_start+G]; Calculate the difference between the actual start timestamp of the instance and the start time T_start, and use it as the boundary offset Offset; Construct a time interval line segment Line={lnterval,Offset} carrying boundary offsets; Existing cloud billing systems only record billing amounts or start and end times, without paying attention to billing boundary offsets caused by non-hourly startups. This rule leads to a certain deviation between the actual runtime of an instance and the billing window, and this deviation cannot achieve fine-grained resource reuse across instances in the time dimension. Introducing boundary offset parameters can quantify discrete billing information, preparing for subsequent fine-grained resource reuse across instances. The overlapping area of ​​phase difference appearing in different instance billing windows is identified by interval algebra operations. The specific operation is as follows: From the established set of time interval segments, extract the billing window end sub-interval for each product instance one by one, and combine the end sub-intervals of all product instances into a set of intervals to be processed. The starting point of the terminal sub-interval is the time point corresponding to the preset duration threshold △T that is traced back from the end time of the billing window of this instance, and the ending point is the end time of the billing window. The preset duration threshold △T represents the maximum remaining duration for which the system determines a billing window has entered the end state; Let any one of the end sub-intervals in the set of operation intervals be denoted as the supply-side instance A, and another any one of the end sub-intervals be denoted as the demand-side instance B (the two are different). When the intersection of the end sub-intervals of A and B on the time axis is >0, calculate the first coverage rate of the intersection duration as a percentage of the duration of the end sub-interval of the supply-side instance A, and the second coverage rate as a percentage of the duration of the end sub-interval of the demand-side instance B, respectively. If the first coverage rate is greater than or equal to the preset first threshold value and the second coverage rate is greater than or equal to the preset second threshold value, then the intersection is marked as the phase difference overlap region, and the start and end timestamps, duration, and identifiers of the supplier instance A and demand instance B of the overlap region are recorded. The specific steps for constructing a cross-instance billing phase correlation graph are as follows: All the marked phase difference overlap areas are used as edge sets, and the supply-side instance and demand-side instance recorded in each phase difference overlap area are used as nodes to construct a cross-instance billing phase association graph. Each edge of the cross-instance billing phase association graph is represented by a directed edge pointing from the supply-side instance node to the demand-side instance node corresponding to the overlapping area, and an edge attribute set is attached. The set of edge attributes includes: the start and end timestamps of the overlapping area, the duration of the overlap, the first coverage rate, and the idle computing power specifications that the supplier instance can release within the overlapping area. When the billing window of any product instance changes, a partial update of the cross-instance billing phase correlation graph is triggered, with the specific rules as follows: If the last sub-interval of the billing window corresponding to the product instance has slipped out of the instance at the current time, then remove all edges with that instance as the supply-side node or demand-side node. If a new phase difference overlap region is identified based on the end sub-interval of the billing window corresponding to the product instance, then a directed edge is added between the nodes of the product instance and the edge attribute set is filled.

[0019] Step 2: Monitor the instantaneous resource demand queue length of demand-side instance B. When the queue increase exceeds the preset expansion threshold, query all supply-side instances that have an associated edge with instance B from the cross-instance billing phase association graph. In a cloud-native environment, the bottleneck of a product instance's business capacity is usually manifested as the accumulation of request queues, rather than the exhaustion of computing resources; For example, when a sudden surge in traffic arrives for a web service, the request will first enter the SYN queue of the listening socket or the thread pool queue of the application layer. At this time, the CPU utilization may still be within the normal range, but the response latency perceived by the client has already begun to rise. If the expansion is triggered only after the CPU utilization exceeds the threshold, the service quality will have been damaged. Therefore, the queue increase is chosen as the trigger signal so that the system can detect the expansion requirement in advance before the resource indicators deteriorate. For each supplier instance, determine whether the start and end timestamps of the overlapping area recorded in the associated edges cover the current time. If they do, extract the supplier instance A corresponding to that associated edge, and calculate the remaining effective duration and available idle computing power of instance A within the overlapping area. The specific operation is as follows: Extract the start and end timestamps of the overlapping area between supplier instance A and demand instance B from the cross-instance billing phase correlation graph, and use the difference between the end time T_end of the overlapping area and the current physical time T_current as the base remaining duration △t_base. The base remaining time △t_base is corrected using the boundary offset Offset of the supplier instance A, and the effective remaining time △t_valid=min(△t_base,GA-Offset-△t_elapsed) is calculated, where GA is the billing granularity unit of supplier instance A, and △t_elapsed is the total running time of supplier instance A since the start of the current billing window. Get the average CPU utilization U_cpu and average memory usage U_mem of supplier instance A within the preset sampling window before the current time, and get the specification configuration of the instance from the instance metadata, including the number of vCPU cores C_total and the total memory M_total; Calculate the number of freed idle vCPU cores C_idle=C_total×(1-U_cpu)×ɑ, and the freed idle memory capacity M_idle=M_total×(1-U_mem)×β, where ɑ and β are preset borrowable scaling factors; The specification for releasable idle computing power is Spec={vCPU:C_idle, memory:M_idle}; At the same time, it is matched and verified with the amount of temporary expansion resources required by instance B. The specific steps are as follows: Using the available idle computing power specification A_Spec and effective remaining duration A_△t_valid of supplier instance A as input, we perform three-dimensional verification against the temporary expansion resource requirement B_Spec and expansion duration requirement B_△t_valid of demand instance B: Dimension 1: Verify whether the resource specifications are met, i.e., A_C_idle≥B_C_idle and A_M_idle≥B_M_idle; Dimension 2: Whether the verification time meets the condition that A_△t_valid ≥ B_△t_valid; Dimension three: Verify whether the feasibility of the secondment is met, namely... Where △t_migrate is the execution time required for the borrowing operation. This is the preset tolerance coefficient for the secondment execution time, with a value range of (0,1). The supplier instance A that simultaneously meets the above three verification conditions is marked as a candidate supplier instance; For all marked candidate supply instances, calculate the comprehensive matching score for each candidate instance. The specific operation is as follows: Calculate the resource specification matching degree S_spec, duration coverage S_time, and billing sunk value S_cost for each candidate instance; The resource specification matching degree S_spec is taken as the smaller value of the ratio between the supply specification and the demand specification in the two dimensions of vCPU cores and memory capacity, that is, S_spec=min(A_C_idle / B_C_idle,A_M_idle / B_M_idle); The duration coverage rate S_time is the ratio of the effective remaining duration to the required duration, i.e., S_time = A_△t_valid / B_△t_valid; The sunk cost of the billing is the ratio of the duration of overlap to the billing granularity of the supply instance. Calculate the comprehensive matching score Grade for each candidate instance: Grade = q1 × S_spec + q2 × S_time + q3 × S_cost, where q1, q2, and q3 are preset weights that satisfy q1 + q2 + q3 = 1 and q1 > q2 > q3. Select the supplier instance A with the highest overall matching score as the most suitable supplier instance A1.

[0020] Step 3: Issue a memory hot migration command to the control plane of the host machine where the most suitable supplier instance A1 is located. Because the host control plane has direct read and write permissions to the local instance memory pages and vCPU context, it can transfer computing power between physical hosts without the cloud provider's billing system being aware of it, thus without incurring any new costs at the billing level. Hot migration technology allows memory pages to be iteratively copied from the source host to the target host while the virtual machine or container is still running. If cold migration is used, A1 must be suspended first, and then the memory image is copied to the B side. A1 completely stops service during this period. However, A1 still needs to handle its original non-borrowed port traffic during the borrowing period. The service interruption of seconds to minutes caused by cold migration is unacceptable. And asynchronously copy the memory page A1 to the reserved container slot in the resource pool where B is located; The reserved container slot refers to a container placeholder that is pre-created in the resource pool where B is located but has not been assigned a specific workload. It has a network namespace, storage mount point and basic system image, but no business code is injected. When the asynchronous copy of A1's memory page arrives, the system directly maps the memory page to the process address space of the slot, so that the slot instantly inherits A1's execution context and computing power. At the same time, the virtual network forwarding plane is updated to remap some traffic addresses pointing to A1 to temporary slots in B, completing the cross-product borrowing and deployment of computing power within the billing phase difference window; The specific rules for remapping a portion of the traffic address pointing to A1 to the temporary slot of B are as follows: Extract the set of target traffic addresses that conform to the preset traffic identification rules from the virtual network forwarding plane of A1; The traffic identification rules include: The target IP address is a virtual IP of A1 and the target port belongs to the preset set of available ports; By strictly limiting the target IP to A1's dedicated virtual IP, the system can create a routing rule in the network forwarding plane that only applies to traffic flowing to A1, without affecting traffic pointing to other backend instances. The source IP address belongs to the preset tenant network segment range; Identify which tenant's inactivity caused the idle computing power and associate the benefits or risks of borrowing with the specific tenant; The HTTP header carries a predefined tenant identifier; The traffic identification component needs to perform deep parsing of the HTTP request header when traffic reaches the load balancer or Ingress gateway, extract the preset tenant identifier field and compare it with the configured list of available tenants: successfully matched requests are marked and included in the remapping scope, while unmatched requests are forwarded along the original path. In multi-product linkage scenarios, even if the supply-side instance A1 and the demand-side instance B are located in different availability zones or different virtual private clouds, as long as the tenant identifier in the HTTP header is consistent, the system can complete accurate traffic filtering and migration. The next-hop address in the forwarding routing table entry corresponding to the target traffic address set is updated to the network address of B's ​​temporary slot, so that the matching traffic that subsequently arrives at the virtual network forwarding plane is forwarded to B's temporary slot.

[0021] As a third embodiment of this application, such as Figure 2 As shown, the present invention provides a dynamic cloud resource deployment system that supports intelligent linkage of multiple products, comprising: The graph construction module collects the start timestamp and billing granularity unit of the current billing window of the cloud account where each product instance is located, maps the billing window of each instance to a time interval line segment with boundary offset, and marks the phase difference overlap area at the end of the billing window of different instances through interval algebra operations, and constructs a cross-instance billing phase correlation graph. The instance filtering module monitors the instantaneous resource demand queue length of demand-side instance B. When the queue increase exceeds the preset expansion threshold, it queries all supply-side instances that have an associated edge with instance B from the cross-instance billing phase association graph. It then checks whether the start and end timestamps of the overlapping area recorded in the associated edge of each supply-side instance cover the current time. If they cover, it extracts the supply-side instance A corresponding to the associated edge, calculates the remaining effective duration and available idle computing power specification of instance A in the overlapping area, and matches and verifies them with the amount of temporary expansion resources required by instance B to filter out the most suitable supply-side instance A1. The temporary scheduling module issues a memory hot migration instruction to the control plane of the host machine where the most suitable supplier instance A1 is located, and asynchronously copies the memory pages of A1 to the reserved container slots in the resource pool where B is located. At the same time, it updates the virtual network forwarding plane to remap some traffic addresses pointing to A1 to the temporary slots of B, thus completing the cross-product borrowing and deployment of computing power within the billing phase difference window.

[0022] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0023] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for dynamically deploying cloud resources that supports intelligent linkage among multiple products, characterized in that: include: Step 1: Collect the start timestamp and billing granularity unit of the current billing window of the cloud account where each product instance is located, map the billing window of each instance to a time interval line segment with boundary offset, and mark the phase difference overlap area at the end of the billing window of different instances to construct a cross-instance billing phase correlation map. Step 2: Monitor the instantaneous resource demand queue length of demand-side instance B. When the queue increase exceeds the preset expansion threshold, query all supply-side instances that have an associated edge with instance B from the cross-instance billing phase association graph, and determine whether the start and end timestamps of the overlapping area recorded in the associated edge of each supply-side instance cover the current time. If they cover, extract the supply-side instance A corresponding to the associated edge, calculate the remaining effective duration and the available idle computing power specification of instance A in the overlapping area, and match and verify it with the amount of temporary expansion resources required by instance B to select the most suitable supply-side instance A1. Step 3: Issue a memory hot migration command to the control plane of the host machine where the most suitable supplier instance A1 is located, and asynchronously copy the memory pages of A1 to the reserved container slots of the resource pool where B is located. At the same time, update the virtual network forwarding plane to remap some traffic addresses pointing to A1 to temporary slots of B, and complete the cross-product borrowing and deployment of computing power within the billing phase difference window.

2. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 1, characterized in that, The specific operation of mapping each instance's billing window to a time interval segment with boundary offsets is as follows: Obtain the billing granularity unit G and the start time T_start of the current billing window for the instance from the instance metadata service or metering API interface provided by the cloud vendor; With T_start as the reference origin, the billing window of this instance is represented as a closed interval lnterval=[T_start,T_start+G]; Calculate the difference between the actual start timestamp of the instance and the start time T_start, and use it as the boundary offset Offset; Construct a time interval line segment Line={lnterval,Offset} carrying boundary offsets.

3. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 1, characterized in that, The specific steps for identifying the overlapping area of ​​phase difference at the end of the billing window for different instances are as follows: From the established set of time interval segments, extract the billing window end sub-interval for each product instance one by one, and combine the end sub-intervals of all product instances into a set of intervals to be processed. The starting point of the terminal sub-interval is the time point corresponding to the preset duration threshold △T that is traced back from the end time of the billing window of this instance, and the ending point is the end time of the billing window. The preset duration threshold △T represents the maximum remaining duration for which the system determines a billing window has entered the end state; Let any one of the end sub-intervals in the set of operation intervals be denoted as the supply-side instance A, and another any one of the end sub-intervals be denoted as the demand-side instance B. When the intersection of the end sub-intervals of A and B on the time axis is >0, calculate the first coverage rate of the intersection duration as a percentage of the duration of the end sub-interval of the supply-side instance A, and the second coverage rate as a percentage of the duration of the end sub-interval of the demand-side instance B. If the first coverage rate is greater than or equal to the preset first threshold value, and the second coverage rate is greater than or equal to the preset second threshold value, then the intersection is marked as a phase difference overlap region, and the start and end timestamps, duration, and identifiers of the supplier instance A and demand instance B of the overlap region are recorded.

4. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 1, characterized in that, The specific steps for constructing a cross-instance billing phase correlation graph are as follows: All the marked phase difference overlap areas are used as edge sets, and the supply-side instance and demand-side instance recorded in each phase difference overlap area are used as nodes to construct a cross-instance billing phase association graph. Each edge of the cross-instance billing phase association graph is represented by a directed edge pointing from the supply-side instance node to the demand-side instance node corresponding to the overlapping area, and an edge attribute set is attached. The set of edge attributes includes: the start and end timestamps of the overlapping area, the duration of the overlap, the first coverage rate, and the idle computing power specifications that the supplier instance can release within the overlapping area. When the billing window of any product instance changes, a partial update of the cross-instance billing phase association graph is triggered.

5. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 4, characterized in that, The specific rules for triggering a partial update of the cross-instance billing phase correlation graph are as follows: If the last sub-interval of the billing window corresponding to the product instance has slipped out of the instance at the current time, then remove all edges with that instance as the supply-side node or demand-side node. If a new phase difference overlap region is identified based on the end sub-interval of the billing window corresponding to the product instance, then a directed edge is added between the nodes of the product instance and the edge attribute set is filled.

6. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 1, characterized in that, The specific steps for calculating the remaining effective duration and available idle computing power specification of instance A within the overlapping region are as follows: Extract the start and end timestamps of the overlapping area between supplier instance A and demand instance B from the cross-instance billing phase correlation graph, and use the difference between the end time T_end of the overlapping area and the current physical time T_current as the base remaining duration △t_base. The base remaining time △t_base is corrected using the boundary offset Offset of the supplier instance A, and the effective remaining time △t_valid=min(△t_base,GA-Offset-△t_elapsed) is calculated, where GA is the billing granularity unit of supplier instance A, and △t_elapsed is the total running time of supplier instance A since the start of the current billing window. Get the average CPU utilization U_cpu and average memory usage U_mem of supplier instance A within the preset sampling window before the current time, and get the specification configuration of the instance from the instance metadata, including the number of vCPU cores C_total and the total memory M_total; Calculate the number of freed idle vCPU cores C_idle=C_total×(1-U_cpu)×ɑ, and the freed idle memory capacity M_idle=M_total×(1-U_mem)×β, where ɑ and β are preset borrowable scaling factors; The available idle computing power specification is obtained as Spec={vCPU:C_idle, memory:M_idle}.

7. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 1, characterized in that, The specific steps for matching and verifying the amount of temporary expansion resources required by instance B are as follows: Taking the available idle computing power specification A_Spec and the effective remaining duration A_△t_valid of supplier instance A as input, and comparing them with the temporary expansion resource requirement B_Spec and expansion duration requirement B_△t_valid of demand instance B, a three-dimensional verification is performed: Dimension 1: Verify whether the resource specification is met, i.e., A_C_idle ≥ B_C_idle and A_M_idle ≥ B_M_idle; Dimension 2: Verify whether the duration is met, i.e., A_△t_valid ≥ B_△t_valid; Dimension 3: Verify whether the borrowing feasibility is met, i.e. Where △t_migrate is the execution time required for the borrowing operation. This is the preset tolerance coefficient for the secondment execution time; The supplier instance A that simultaneously meets the above three verification conditions is marked as a candidate supplier instance. For all marked candidate supplier instances, the comprehensive matching score of each candidate instance is calculated, and the supplier instance A with the highest comprehensive matching score is selected as the most suitable supplier instance A1.

8. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 7, characterized in that, The specific steps for calculating the overall matching score for each candidate instance are as follows: Calculate the resource specification matching degree S_spec, duration coverage S_time, and billing sunk value S_cost for each candidate instance; The resource specification matching degree S_spec is taken as the smaller value of the ratio between the supply specification and the demand specification in the two dimensions of vCPU cores and memory capacity, that is, S_spec=min(A_C_idle / B_C_idle,A_M_idle / B_M_idle); The duration coverage rate S_time is the ratio of the effective remaining duration to the required duration, i.e., S_time = A_△t_valid / B_△t_valid; The sunk cost of the billing is the ratio of the duration of overlap to the billing granularity of the supply instance. Calculate the comprehensive matching score Grade for each candidate instance: Grade = q1 × S_spec + q2 × S_time + q3 × S_cost, where q1, q2, and q3 are preset weights that satisfy q1 + q2 + q3 = 1 and q1 > q2 > q3.

9. The cloud resource dynamic deployment method supporting intelligent linkage of multiple products according to claim 1, characterized in that, The specific rules for remapping a portion of the traffic address pointing to A1 to the temporary slot of B are as follows: Extract the set of target traffic addresses that conform to the preset traffic identification rules from the virtual network forwarding plane of A1; The traffic identification rules include: the target IP address is a virtual IP of A1 and the target port belongs to a preset set of available ports; the source IP address belongs to a preset tenant network segment; and the HTTP header carries a preset tenant identifier. The next-hop address in the forwarding routing table entry corresponding to the target traffic address set is updated to the network address of B's ​​temporary slot, so that the matching traffic that subsequently arrives at the virtual network forwarding plane is forwarded to B's temporary slot.

10. A cloud resource dynamic deployment system supporting intelligent linkage of multiple products, used to execute the cloud resource dynamic deployment method supporting intelligent linkage of multiple products as described in any one of claims 1-9, characterized in that, include: The graph construction module collects the start timestamp and billing granularity unit of the current billing window of the cloud account where each product instance is located, maps the billing window of each instance to a time interval line segment with boundary offset, and marks the phase difference overlap area at the end of the billing window of different instances through interval algebra operations, and constructs a cross-instance billing phase correlation graph. The instance filtering module monitors the instantaneous resource demand queue length of demand-side instance B. When the queue increase exceeds the preset expansion threshold, it queries all supply-side instances that have an associated edge with instance B from the cross-instance billing phase association graph. It then checks whether the start and end timestamps of the overlapping area recorded in the associated edge of each supply-side instance cover the current time. If they cover, it extracts the supply-side instance A corresponding to the associated edge, calculates the remaining effective duration and available idle computing power specification of instance A in the overlapping area, and matches and verifies them with the amount of temporary expansion resources required by instance B to filter out the most suitable supply-side instance A1. The temporary scheduling module issues a memory hot migration instruction to the control plane of the host machine where the most suitable supplier instance A1 is located, and asynchronously copies the memory pages of A1 to the reserved container slots in the resource pool where B is located. At the same time, it updates the virtual network forwarding plane to remap some traffic addresses pointing to A1 to the temporary slots of B, thus completing the cross-product borrowing and deployment of computing power within the billing phase difference window.

Images