A decoupled computing system and method based on distributed computing power scheduling

Abstract

This invention discloses a decoupled computing system and method based on distributed computing power scheduling, belonging to the field of distributed computing technology. The system includes: constructing a hierarchical computing power domain structure; long-cycle decision units generating reserved resource status tables for each computing power domain; short-cycle decision units monitoring terminal tasks and marking sets of tasks requiring cross-domain collaboration; a collaborative control center receiving the reserved resource status tables and sets of tasks requiring cross-domain collaboration, publishing a heat index and historical fulfillment rates for each computing power domain at virtual boundary nodes; and short-cycle decision units in each computing power domain submitting bids and generating temporary resource trading pairs, along with associated bids and reputation data. This invention, through a hierarchical decoupling, dynamic game theory, and negotiation confirmation collaborative mechanism, effectively reduces the scheduling pressure on the central node, protects the resource privacy of each computing power domain, improves the accuracy of cross-domain resource matching and the trust foundation for inter-domain interactions, and enhances the global resource utilization and task execution efficiency in multi-computing power domain scenarios.

Description

Technical Field

[0001] This invention relates to the field of distributed computing technology, and in particular to a decoupled computing system and method based on distributed computing power scheduling. Background Technology

[0002] With the rapid development of industrial cloud computing and edge computing technologies, distributed computing power scheduling has become a key technical means to support the collaborative processing of large-scale computing tasks. Current mainstream distributed computing power scheduling methods typically employ a centralized resource management architecture, where a central node collects the status of computing resources across the entire domain and uniformly allocates tasks; or they adopt a decentralized distributed scheduling strategy, where each computing node independently decides the task distribution path based on its local load information. These methods, to a certain extent, realize cross-domain collaboration of computing resources, providing a basic framework for the collaborative execution of multi-regional computing tasks.

[0003] However, existing distributed computing power scheduling methods still have several technical limitations in multi-computing-domain collaborative scenarios. On the one hand, centralized scheduling architectures struggle to cope with the communication overhead and central node processing pressure brought about by the expansion of computing domain scale, and the resource status information of each computing domain needs to be fully exposed to the central node, resulting in insufficient protection of resource privacy within the domain. On the other hand, existing distributed scheduling strategies lack a dynamic perception mechanism for cross-domain resource supply and demand relationships. Each computing domain makes decisions based solely on local load information, making it difficult to form a globally optimized resource matching scheme, resulting in limited computing resource utilization and large fluctuations in cross-domain task response latency. Summary of the Invention

[0004] In view of the aforementioned existing problems, this invention provides a decoupled computing method based on distributed computing power scheduling to solve the problem of insufficient global optimization efficiency caused by the difficulty in dynamically matching resource supply and demand and the lack of trust foundation for inter-domain interaction.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides a decoupled computing method based on distributed computing power scheduling, comprising, A hierarchical computing power domain structure is constructed. Long-cycle decision-making units generate reserved resource status tables for each computing power domain, while short-cycle decision-making units monitor terminal tasks and mark sets of task requests that require cross-domain collaboration. The collaborative control center receives the reserved resource status table and the set of cross-domain collaborative task requests, publishes the heat index and the historical performance rate of each computing power domain at the virtual boundary node, and generates a set of temporary resource trading pairs and associated quotations and credit data for each computing power domain's short-cycle decision-making unit; The long-term decision-making unit of the participating computing power domain receives the set of temporary resource trading pairs and associated price and credit data, compares them with the original reserved resource status table of the domain to calculate the degree of impact, and generates the final cross-domain resource trading agreement and the updated reserved resource status table after negotiation. The short-cycle decision-making unit receives the final cross-domain resource transaction agreement, distributes and executes tasks according to the updated reserved resource status table, records the actual consumption and updates the local reputation data. Each computing power domain reports the actual consumption and the updated local reputation data to the collaborative control center to generate the next round of popularity index and global reputation data.

[0007] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the construction of the computing power domain hierarchical structure includes the following steps: Obtain the physical location coordinates and network topology connections of each computing power hub, and divide computing power hubs that are physically adjacent and have close network connections into the same computing power domain; Long-cycle decision-making units and short-cycle decision-making units are deployed in each computing power domain, and virtual boundary nodes are set between adjacent computing power domains to represent the resource interaction relationship between adjacent computing power domains.

[0008] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the long-cycle decision unit generates a reserved resource status table for each computing power domain, and the short-cycle decision unit monitors terminal tasks and marks the set of task requests requiring cross-domain collaboration, including the following steps. The long-cycle decision-making unit collects historical data on the resource reserves of each computing node in the domain at preset time intervals, predicts future resource demand based on the historical data on resource reserves, and converts it into a reserved resource status table for the domain. The short-cycle decision unit receives terminal task requests in real time, analyzes computing power requirements and task volume, and matches them with the reserved resource status table of this domain. When there are sufficient available resources in this domain and the reserved resources are not occupied, the terminal task request is executed within this domain; otherwise, the terminal task request is marked as a cross-domain collaborative task request and a set of cross-domain collaborative task requests is generated.

[0009] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the generation of the heat index and the historical fulfillment rate of each computing power domain includes the following steps: The collaborative control center receives real-time load, idle resource quantity, actual consumption in the previous cycle, and local reputation data reported by each computing power domain; For each virtual boundary node, calculate the resource supply-demand ratio for the real-time load and idle resources of the associated adjacent computing power domains, and determine the heat index based on the resource supply-demand ratio. The historical fulfillment rate of each computing power domain is calculated based on the degree of matching between the actual consumption in the previous cycle and the agreed resource quantity in the corresponding final cross-domain resource transaction agreement.

[0010] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the generation of the quotation includes the following steps: Each computing power domain's short-cycle decision-making unit acquires the heat index, historical fulfillment rate, and real-time load of its domain, and calculates the idle cost of its domain's resources; Expected revenue is calculated based on popularity index, historical fulfillment rate, real-time load of the domain, and resource idle cost; The type and quantity of computing power that can be released are determined based on expected returns, and pricing information is generated by combining the priority weights corresponding to historical performance rates.

[0011] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the step of generating a set of temporary resource trading pairs and associated price and reputation data includes the following steps: Virtual boundary nodes collect bidding information submitted by adjacent computing power domains and parse the type and quantity of releaseable computing power in each bidding information; The available computing power type is matched with the computing power requirements of the task request set that requires cross-domain collaboration, and paired in order of priority to generate temporary resource transaction pairs. Each temporary resource transaction pair includes the initiating domain, the receiving domain, the transaction resource type, and the transaction resource quantity. Each temporary resource trading pair is associated with its corresponding price information and the reputation data of the initiating domain, generating a set of temporary resource trading pairs and associated price and reputation data.

[0012] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the degree of computational impact includes the following steps: Extract transaction items related to this domain and associated quotation information and reputation data from the temporary resource transaction pair set, and read the original reserved resource status table of this domain. For each transaction, calculate the proportion of the transaction resource quantity to the total resource quantity of the domain during the same period, and determine the impact value based on the proportion; The impact value is compared with the preset fluctuation tolerance range. If the impact value is lower than the lower limit of the tolerance range, it is marked as low impact; if it is between the tolerance range, it is marked as medium impact; and if it is higher than the upper limit of the tolerance range, it is marked as high impact. Based on the marking results, the confirmation status of each transaction in this domain is generated. The confirmation status of low-impact transactions is "confirmed", the confirmation status of high-impact transactions is "needs correction", and the confirmation status of medium-impact transactions is determined based on the comparison between the associated reputation data and the preset reputation threshold.

[0013] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the step of generating the final cross-domain resource transaction agreement and the updated reserved resource status table includes the following steps: Collect the confirmation status and correction requests submitted by each transaction domain for each transaction item. For transaction items with inconsistent confirmation status, send the correction request of the transaction item to the corresponding transaction domain at the virtual boundary node. Each transaction domain adjusts the quotation information according to the correction request and resubmits the confirmation status until the confirmation status of the transaction items is consistent, thus forming the final cross-domain resource transaction agreement. The long-term decision-making units of each transaction domain adjust the reserved resource status table of their domain according to the number of transaction resources confirmed in the final cross-domain resource transaction agreement, and generate an updated reserved resource status table.

[0014] As a preferred embodiment of the decoupled computing method based on distributed computing power scheduling described in this invention, the step of distributing and executing tasks according to the updated reserved resource status table includes the following steps: The short-cycle decision-making unit obtains the final cross-domain resource transaction agreement from the collaborative control center and the updated reserved resource status table from the long-cycle decision-making unit; When a new terminal task request arrives, the computing power requirement of the terminal task request is parsed, and the available resources in this domain are queried in the updated reserved resource status table. When the available resources in this domain are sufficient and the reservation constraints in the updated reserved resource status table are not violated, the terminal task request will be distributed to the computing power nodes in this domain for execution; otherwise, according to the receiving domain determined by the final cross-domain resource transaction protocol, the terminal task request will be forwarded to the virtual boundary node corresponding to the receiving domain for execution. After execution, the execution result and actual consumption are returned to the initiating domain, the actual consumption is recorded, and the local reputation data of this domain is updated according to the performance of the final cross-domain resource transaction agreement.

[0015] Secondly, the present invention provides a decoupled computing system based on distributed computing power scheduling, comprising: The computing power domain construction module builds a hierarchical structure of computing power domains. The long-cycle decision unit generates a reserved resource status table for each computing power domain, and the short-cycle decision unit monitors terminal tasks and marks the set of task requests that require cross-domain collaboration. The collaborative pricing module receives the reserved resource status table and the set of cross-domain collaborative task requests from the collaborative control center. It publishes the heat index and the historical performance rate of each computing power domain at the virtual boundary node. The short-cycle decision-making units of each computing power domain submit quotations and generate a set of temporary resource trading pairs, as well as associated quotations and credit data. The protocol negotiation module receives a set of temporary resource trading pairs and associated price and credit data from the long-term decision-making unit of the transaction computing power domain. It compares the data with the original reserved resource status table of the domain to calculate the degree of impact. After negotiation, it generates the final cross-domain resource trading agreement and the updated reserved resource status table. The execution feedback module receives the final cross-domain resource transaction agreement, distributes and executes tasks according to the updated reserved resource status table, records the actual consumption and updates the local reputation data. Each computing power domain reports the actual consumption and updated local reputation data to the collaborative control center to generate the next round of popularity index and global reputation data.

[0016] The beneficial effects of this invention are as follows: By constructing a hierarchical structure of computing power domains and setting up two-level decision-making units of long-term and short-term cycles, the decoupling of local autonomy and cross-domain collaboration of computing power resources is achieved; the long-term decision-making unit predicts future resource demand based on historical data and generates a reserved resource status table, providing a predictable resource constraint basis for cross-domain collaboration; the short-term decision-making unit monitors terminal tasks in real time and matches them with the reserved resource status table, accurately identifying the set of task requests requiring cross-domain collaboration; the collaborative control center publishes a heat index and historical performance rate at virtual boundary nodes, establishing a dynamic perception mechanism for the supply and demand relationship of resources between domains; each computing power domain independently quotes prices based on expected returns and generates temporary resource trading pairs, introducing a market game mechanism into the computing power scheduling process; the long-term decision-making units participating in the trading domain calculate the impact of trading items on local resource stability and conduct multiple rounds of negotiation to form a final cross-domain resource trading agreement that takes into account both global optimization and local autonomy; the short-term decision-making unit executes task distribution according to the updated reserved resource status table and feeds back the actual consumption and performance status to the collaborative control center, forming a closed-loop update of heat index and reputation data. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a decoupled computing method based on distributed computing power scheduling.

[0019] Figure 2 A flowchart for constructing the computing power domain and generating reserved / tagged tasks.

[0020] Figure 3 Flowchart for coordinated pricing and generation of temporary trading pairs.

[0021] Figure 4A flowchart for generating the final agreement through negotiation. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a decoupled computing method based on distributed computing power scheduling, including the following steps: S1. Construct a hierarchical structure of computing power domains. Long-cycle decision-making units generate reserved resource status tables for each computing power domain, while short-cycle decision-making units monitor terminal tasks and mark sets of task requests that require cross-domain collaboration.

[0026] S1.1. By using the network topology discovery protocol and node location service deployed in the global computing power network, the physical location coordinates of each computing power hub and the network topology connection relationship between each computing power hub are automatically collected; based on the obtained physical location coordinates and network topology connection relationship, computing power hubs that are physically adjacent and have close network connections are divided into the same computing power domain, and each computing power domain operates independently as an autonomous unit, thus completing the division of computing power domains.

[0027] Within each computing power domain, long-cycle decision-making units and short-cycle decision-making units are deployed. Long-cycle decision-making units operate at preset time intervals, determined through statistical analysis based on historical task arrival fluctuation cycles. These intervals are typically set as integer multiples of the minimum cycle length for task arrival rate changes, such as 24 hours or 1 hour, to ensure the ability to capture cyclical changes in resource demand. Short-cycle decision-making units operate continuously in real-time. A virtual boundary node is established between two adjacent computing power domains to represent the resource interaction relationships between them. The resulting computing power domains, the long-cycle and short-cycle decision-making units deployed within each domain, and the virtual boundary node between adjacent computing power domains together constitute a hierarchical computing power domain structure.

[0028] After completing the construction of the hierarchical structure of computing power domains, the long-cycle decision-making units and short-cycle decision-making units within each computing power domain are activated. The long-cycle decision-making units collect historical data on the resource reserves of all computing power nodes within their respective computing power domains at multiple points in the past, according to preset time intervals.

[0029] Suppose a certain computing node in the past... The remaining resources at each point in time are denoted as follows: The number of time points collected is Based on historical resource surplus data, the long-term decision-making unit predicts resource demand for the next time period, expressed as: ; in, This represents the projected value of resource demand. Indicates the computing power node in the past [number]th Historical data on resource availability at each point in time. Indicates the relationship with the first The weight coefficients for each time point are determined using an exponential decay method based on their distance from the current time point. The closer a time point is to the current time point, the larger its weight coefficient; the farther a time point is from the current time point, the smaller its weight coefficient. This represents the total number of historical data points on the collected resource reserves.

[0030] S1.2. Based on the calculated resource demand forecast, the long-term decision unit determines the types and quantities of resources that the computing power domain needs to reserve for future time periods, and organizes the types and quantities of reserved resources into a reserved resource status table for the computing power domain. The reserved resource status table includes the types of resources that each computing power node in the computing power domain needs to reserve and the corresponding reserved resource quantities for different time periods.

[0031] The short-cycle decision unit operates in a continuous real-time manner, continuously receiving terminal task requests from terminal devices. When a terminal task request is received, the short-cycle decision unit parses the corresponding terminal task request to obtain the computing power type and task quantity required by the terminal task request.

[0032] Meanwhile, the short-cycle decision unit obtains the current valid reserved resource status table of this computing power domain from the long-cycle decision unit, and matches the computing power resources required by the corresponding terminal task request with the reserved resource information in the reserved resource status table.

[0033] During the matching process, the short-cycle decision unit first checks whether the total amount of computing power resources currently available in this computing power domain meets the computing power requirements of the corresponding terminal task request, and then checks whether using the resources of this domain to execute the corresponding task will occupy the resources already reserved in the reserved resource status table.

[0034] When there are sufficient available resources in the computing power domain and executing the corresponding task will not consume the reserved resources, the short-cycle decision unit marks the corresponding terminal task request as to be executed within the domain and distributes the task to the computing power nodes within the domain for processing.

[0035] When the available resources in this computing domain are insufficient or executing the corresponding task will consume the reserved resources, the short-cycle decision unit marks the corresponding terminal task request as a cross-domain collaborative task request and summarizes all marked cross-domain collaborative task requests to generate a cross-domain collaborative task request set.

[0036] S2. The Collaborative Control Center receives the reserved resource status table and the set of cross-domain collaborative task requests, publishes the heat index and historical performance rate of each computing power domain at the virtual boundary node, and generates a set of temporary resource trading pairs and associated quotes and reputation data for each computing power domain's short-cycle decision-making unit.

[0037] S2.1. The Collaborative Control Center receives periodic reports from each computing power domain on real-time load, idle resources, actual consumption in the previous cycle, and reputation information maintained locally by each computing power domain. Real-time load represents the total resources used by all currently executing tasks within the current computing power domain; idle resources represent the total amount of currently unused available resources within the current computing power domain; actual consumption in the previous cycle represents the amount of resources actually used by the current computing power domain to execute cross-domain tasks in the previous complete operating cycle; and local reputation information represents the subjective evaluation data of the current computing power domain on the performance of other computing power domains based on historical interaction records.

[0038] For each virtual boundary node, the collaborative control center obtains the real-time load and idle resource quantity of the two adjacent computing power domains associated with the current virtual boundary node, and calculates the resource supply and demand ratio at the current virtual boundary node.

[0039] Set virtual boundary nodes The associated adjacent computing power domains are domains. Heyu The real-time loads are respectively and The amount of idle resources are respectively and Then the formula for calculating the resource supply-demand ratio is: ; Wherein, virtual boundary nodes are represented. The resource supply and demand ratio and Representing fields respectively Heyu Real-time load, and Representing fields respectively Heyu The amount of idle resources.

[0040] The collaborative control center maps the calculated resource supply-demand ratio to a heat index. The heat index is positively correlated with the resource supply-demand ratio; the higher the resource supply-demand ratio, the higher the heat index.

[0041] S2.2. The Collaborative Control Center calculates the historical fulfillment rate of each computing power domain based on the actual consumption reported by each computing power domain in the previous cycle, combined with the agreed resource quantity in the final cross-domain resource transaction agreement formed in the previous cycle.

[0042] For computing power domain Let the actual consumption in the previous cycle be... The corresponding final cross-domain resource transaction agreement stipulates that the amount of resources to be provided is The historical performance rate The calculation method is as follows: like ,but 1; if ,but .

[0043] Historical fulfillment rate reflects computing power domain The extent to which the agreement was fulfilled in the previous cycle is an important component of the computing power domain's reputation information.

[0044] S2.3. The collaborative control center publishes the heat index corresponding to each virtual boundary node and the historical fulfillment rate of the adjacent computing power domain associated with the virtual boundary node on the corresponding virtual boundary node, so that the short-cycle decision-making units of the adjacent computing power domain can obtain them.

[0045] Each computing power domain's short-cycle decision-making unit obtains the popularity index published by the virtual boundary nodes adjacent to its domain, as well as the historical fulfillment rate of adjacent computing power domains, from these virtual boundary nodes. Simultaneously, the short-cycle decision-making unit obtains the real-time load of its domain at the current moment and calculates the idle resource cost based on the current amount of idle resources in its domain. The idle resource cost is directly proportional to the amount of idle resources; the larger the amount of idle resources, the higher the idle resource cost.

[0046] S2.4. Each computing power domain's short-cycle decision-making unit calculates its expected benefit from participating in cross-domain collaboration based on the acquired heat index, the historical fulfillment rate of adjacent computing power domains, the real-time load of its own domain, and the idle resource cost of its own domain. The expected benefit is positively correlated with the heat index and historical fulfillment rate, and negatively correlated with the real-time load and idle resource cost of its own domain. Higher heat index, higher historical fulfillment rates of adjacent domains, lower real-time load of the own domain, and lower idle resource cost of the own domain result in higher expected benefits. The expression for calculating the expected benefit is as follows: ; in, Indicates expected return, Indicates from virtual boundary node The popularity index obtained This represents the historical fulfillment rate of adjacent computing power domains. This indicates the real-time load of this domain at the current moment. This indicates the current amount of idle resources in this domain (i.e., the cost of idle resources).

[0047] Each computing power domain's short-cycle decision-making unit determines the type and quantity of computing power that can be released within the current cycle based on the calculated expected return. Higher expected returns result in a larger quantity of released computing power. Simultaneously, the short-cycle decision-making unit determines the priority weight of the current bid based on the historical fulfillment rate of adjacent computing power domains; adjacent domains with higher historical fulfillment rates have greater priority weights. The short-cycle decision-making unit encapsulates the type and quantity of releaseable computing power, the expected resource exchange ratio, and the priority weights into bidding information and submits this information to the virtual boundary node adjacent to its domain.

[0048] S2.5. The virtual boundary node collects the quotation information submitted by all adjacent computing power domains and parses the type and quantity of releaseable computing power in each quotation. At the same time, the virtual boundary node obtains the set of task requests that require cross-domain collaboration, which includes all task requests marked as requiring cross-domain execution. Each task request includes the type and quantity of computing power required for the task.

[0049] Virtual boundary nodes match the types of releaseable computing power with the computing power requirements of tasks in the set of cross-domain collaborative task requests. For each task request, bids whose releaseable computing power type matches the task's computing power requirement are selected from the bid information and matched in descending order of priority weight in the bid information. During the matching process, each temporary resource transaction pair consists of an initiating domain (the computing power domain providing the resource), a receiving domain (the computing power domain requiring the resource), a transaction resource type, and a transaction resource quantity. The transaction resource quantity is the smaller value between the releaseable computing power quantity in the bid and the task quantity in the task request, and is deducted accordingly from both the releaseable computing power quantity in the bid and the task quantity in the task request. This process is repeated until all task requests are matched or all releaseable computing power in all bids is allocated.

[0050] The virtual boundary node will associate each generated temporary resource trading pair with the pricing information of the initiating domain and the reputation information of the initiating domain (i.e., the historical performance rate of the initiating domain), forming a set of temporary resource trading pairs and associated pricing and reputation information.

[0051] S3. The long-term decision-making unit of the participating computing power domain receives the set of temporary resource trading pairs and associated price and credit data, compares them with the original reserved resource status table of this domain to calculate the degree of impact, and generates the final cross-domain resource trading agreement and the updated reserved resource status table after negotiation.

[0052] S3.1. The long-term decision-making unit participating in the computing power domain receives the set of temporary resource trading pairs and associated price and reputation data; it extracts all trading items involving its domain from the set of temporary resource trading pairs. Each trading item includes the initiating domain, receiving domain, trading resource type, and trading resource quantity. Simultaneously, it extracts the associated price information and the reputation data of the initiating domain for each trading item. The long-term decision-making unit reads the original reserved resource status table of its domain from local storage. This table records the types of resources and corresponding quantities that the computing power domain needs to reserve at different future time periods.

[0053] For each transaction involving this domain, the long-term decision-making unit calculates the proportion of the transaction resource quantity to the total resource quantity of this domain during the same period, and determines the impact value based on the proportion.

[0054] Let the total resources of this domain during the same period be The number of resources traded is The expression for calculating the impact value is: ; in, This represents the impact of a transaction on the stability of local domain resources, reflecting the relative amount of local domain resources that a transaction may consume. This indicates the number of transaction resources involving this domain in a transaction item. This indicates the total amount of resources in this domain during the same time period.

[0055] S3.2. The long-term decision-making unit compares the calculated impact value with the preset volatility tolerance range. The volatility tolerance range consists of a lower limit and an upper limit, determined based on the standard deviation of historical resource usage data in this domain. For example, the lower limit is 0.5 times the standard deviation of historical resource usage volatility, and the upper limit is 2 times the standard deviation. If the impact value is lower than the lower limit, the transaction is marked as low impact; if the impact value is between the lower and upper limits, it is marked as medium impact; if the impact value is higher than the upper limit, it is marked as high impact.

[0056] The long-term decision-making unit generates the confirmation status of each transaction item within its domain based on the marking results. For transactions marked as low impact, the confirmation status is "Confirmed," indicating that the domain agrees to execute the transaction under the current conditions. For transactions marked as high impact, the confirmation status is "Needs Correction," indicating that the domain disagrees with the current transaction and requires the other party to adjust its quotation information. For transactions marked as medium impact, the confirmation status is further compared with the reputation data of the initiating domain associated with the transaction and a preset reputation threshold. The preset reputation threshold is determined based on global historical transaction statistics, for example, the average historical fulfillment rate of all computing power domains. If the reputation data is greater than or equal to the reputation threshold, the confirmation status is "Confirmed"; if the reputation data is less than or equal to the reputation threshold, the confirmation status is "Needs Correction."

[0057] For transactions whose status is confirmed to require correction, the long-term decision-making unit determines the acceptable range of correction based on the actual situation of resources in this domain, such as the percentage by which the number of transaction resources can be reduced or the acceptable range of replacement ratio adjustment, and includes the range of correction as part of the correction request.

[0058] S3.3. Each long-term decision-making unit in the participating transaction computing power domain sends its confirmation status for all transaction items and correction requests for transaction items requiring correction to the collaborative control center. The collaborative control center collects the confirmation status and correction requests submitted by all transaction domains.

[0059] The collaborative control center checks whether the confirmation status of each transaction item is consistent across all relevant transaction domains. If a transaction item is confirmed in all relevant transaction domains, the transaction item is agreed upon. If there are inconsistencies in the confirmation status, such as one party confirming the transaction while the other party requires correction, or both parties requiring correction but with conflicting correction requests, the collaborative control center marks the transaction item as a conflicting transaction item and initiates a negotiation process at the virtual boundary node corresponding to the transaction item.

[0060] During the negotiation process, the Collaborative Control Center sends the correction request for the conflicting transaction item to the long-term decision-making unit of the corresponding transaction domain. Based on the received correction request, the long-term decision-making unit of the corresponding transaction domain first determines, in conjunction with the domain's original reserved resource status table, the domain's confirmation status for the transaction item, and the associated quotation information, whether the confirmation status can be directly updated without adjusting the quotation information. If the confirmation status can be directly updated without adjusting the quotation information, the long-term decision-making unit regenerates the confirmation status and correction request and submits it to the Collaborative Control Center.

[0061] If the correction request involves adjustments to the amount of releaseable computing power or the expected resource replacement ratio, the long-term decision-making unit of the corresponding transaction domain will feed back the correction information to the short-term decision-making unit of the same domain. The short-term decision-making unit of the same domain will update the quotation information corresponding to the transaction item based on the correction information. The updated quotation information still includes the type of releaseable computing power, the amount of releaseable computing power, the expected resource replacement ratio, and the priority weight. After the short-term decision-making unit returns the updated quotation information to the long-term decision-making unit of the same domain, the long-term decision-making unit of the same domain will then regenerate the confirmation status and correction request based on the updated quotation information and the original reserved resource status table of the same domain, and submit them to the collaborative control center.

[0062] Therefore, the generation and updating of confirmation status is the responsibility of the long-cycle decision-making unit, while the generation and updating of quotation information is the responsibility of the short-cycle decision-making unit. The process of each trading domain adjusting quotation information and resubmitting confirmation status according to correction requests is unfolded as a collaborative process between the long-cycle and short-cycle decision-making units within the same trading domain. This process is repeated until the confirmation status of all trading items in all relevant trading domains is consistent.

[0063] Once all transaction items are agreed upon, the Collaborative Control Center will summarize the finalized transaction items to form a final cross-domain resource transaction agreement. This agreement includes the resource type and the final confirmed quantity of resources for each transaction item. The Collaborative Control Center will then distribute the final cross-domain resource transaction agreement to the long-term decision-making units of each participating computing power domain. Based on the confirmed quantity of resources in the final cross-domain resource transaction agreement, each participating computing power domain's long-term decision-making unit will adjust its original reserved resource status table: for transaction items where the domain is the initiating domain (providing resources), the corresponding quantity of resources will be deducted from the reserved resources; for transaction items where the domain is the receiving domain (using resources), the corresponding quantity of resources will be added to the reserved resources. After the adjustment, an updated reserved resource status table will be generated.

[0064] S4. The short-cycle decision unit receives the final cross-domain resource transaction agreement, distributes and executes tasks according to the updated reserved resource status table, records the actual consumption and updates the local reputation data. Each computing power domain reports the actual consumption and the updated local reputation data to the collaborative control center to generate the next round of heat index and global reputation data.

[0065] S4.1. The short-cycle decision-making unit obtains the final cross-domain resource transaction agreement from the collaborative control center and the updated reserved resource status table from the long-cycle decision-making unit of this computing power domain. The final cross-domain resource transaction agreement contains all agreed-upon transaction items, and each transaction item records the initiating domain, receiving domain, transaction resource type, and the final confirmed quantity of transaction resources. The updated reserved resource status table records the types of reserved resources actually available in this computing power domain at different future time periods and the corresponding quantities of reserved resources.

[0066] The short-cycle decision unit continuously receives terminal task requests from terminal devices. When a new terminal task request arrives, the short-cycle decision unit parses the request to obtain the computing power type and task volume required. The short-cycle decision unit queries the updated reserved resource status table of the computing power domain to determine the total amount of computing power resources available in the domain during the current time period. It also checks whether the available resources meet the computing power requirements of the terminal task request and whether executing the corresponding task will consume the resources already reserved in the updated reserved resource status table.

[0067] If the available resources in this computing power domain are sufficient and the execution of the task will not consume the reserved resources in the updated reserved resource status table, the short-cycle decision unit will mark the terminal task request as to be executed within this domain and distribute the task to the computing power nodes within this domain for processing.

[0068] If the available resources in this computing power domain are insufficient, or if executing the task will occupy the reserved resources in the updated reserved resource status table, the short-cycle decision unit determines the receiving domain according to the final cross-domain resource transaction protocol. Specifically, the short-cycle decision unit searches for transaction items in the final cross-domain resource transaction protocol that can provide the required computing power type, and uses the receiving domain agreed in the transaction item as the target domain for this task forwarding. Then, the terminal task request is forwarded to the virtual boundary node corresponding to the target domain.

[0069] S4.2. When a terminal task request is executed within this domain, upon completion, the computing nodes within this domain return the execution result and the actual resource consumption to the short-cycle decision unit. When a terminal task request is forwarded to a partner domain for execution, after the partner domain completes execution, it returns the execution result and the actual resource consumption to the short-cycle decision unit of the initiating domain via a virtual boundary node. The short-cycle decision unit of the initiating domain receives and records the actual consumption and adds it to the total actual consumption record of this domain.

[0070] The short-cycle decision-making unit of the initiating domain updates its local reputation data for the partner domain based on the performance of this interaction. The performance is measured by the ratio of the actual consumption to the number of transaction resources agreed in the final cross-domain resource transaction agreement. The closer the ratio is to 1, the better the performance. The local reputation data is updated by averaging the original local reputation data with the performance ratio of this interaction to obtain the new local reputation data. The new local reputation data reflects the latest subjective evaluation of the performance of the partner domain by the initiating domain.

[0071] All computing power domains report their actual consumption and updated local reputation data for the current period to the collaborative control center according to a preset time cycle. After receiving the data reported by all computing power domains, the collaborative control center recalculates the historical fulfillment rate of each computing power domain using the same calculation method, based on the actual consumption reported by each domain, and uses this as the global reputation data for the next round. Simultaneously, the collaborative control center recalculates the resource supply-demand ratio of each virtual boundary node using the same calculation method, based on the real-time load and idle resource amounts reported by each computing power domain (this data is reported periodically or obtained at this time), and uses this resource supply-demand ratio as the heat index for the next round.

[0072] This embodiment also provides a decoupled computing system based on distributed computing power scheduling, including: The computing power domain construction module builds a hierarchical structure of computing power domains. The long-cycle decision unit generates a reserved resource status table for each computing power domain, and the short-cycle decision unit monitors terminal tasks and marks the set of task requests that require cross-domain collaboration. The collaborative pricing module receives the reserved resource status table and the set of cross-domain collaborative task requests from the collaborative control center. It publishes the heat index and the historical performance rate of each computing power domain at the virtual boundary node. The short-cycle decision-making units of each computing power domain submit quotations and generate a set of temporary resource trading pairs, as well as associated quotations and credit data. The protocol negotiation module receives a set of temporary resource trading pairs and associated price and credit data from the long-term decision-making unit of the transaction computing power domain. It compares the data with the original reserved resource status table of the domain to calculate the degree of impact. After negotiation, it generates the final cross-domain resource trading agreement and the updated reserved resource status table. The execution feedback module receives the final cross-domain resource transaction agreement, distributes and executes tasks according to the updated reserved resource status table, records the actual consumption and updates the local reputation data. Each computing power domain reports the actual consumption and updated local reputation data to the collaborative control center to generate the next round of popularity index and global reputation data.

[0073] In summary, this invention achieves decoupling of local autonomy and cross-domain collaboration of computing resources by constructing a hierarchical computing power domain structure and setting up long-cycle and short-cycle two-level decision-making units. The long-cycle decision-making unit predicts future resource demand based on historical data and generates a reserved resource status table, providing a predictable resource constraint basis for cross-domain collaboration. The short-cycle decision-making unit monitors terminal tasks in real time and matches them with the reserved resource status table to accurately identify the set of task requests requiring cross-domain collaboration. The collaborative control center publishes a heat index and historical fulfillment rate at virtual boundary nodes to establish a dynamic perception mechanism for the supply and demand relationship of resources between domains. Each computing power domain independently quotes prices based on expected returns and generates temporary resource trading pairs, introducing a market game mechanism into the computing power scheduling process. The long-cycle decision-making units participating in the trading domain calculate the impact of trading items on the stability of local resources and conduct multiple rounds of negotiation to form a final cross-domain resource trading agreement that takes into account both global optimization and local autonomy. The short-cycle decision-making unit executes task distribution according to the updated reserved resource status table and feeds back the actual consumption and fulfillment status to the collaborative control center, forming a closed-loop update of heat index and reputation data.

[0074] It should be noted that the above embodiments are only used to illustrate the technical solutions 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 solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A decoupled computing method based on distributed computing power scheduling, characterized in that: include, A hierarchical computing power domain structure is constructed, which includes each computing power domain, long-cycle decision-making units and short-cycle decision-making units deployed in each computing power domain, and virtual boundary nodes set between adjacent computing power domains; the long-cycle decision-making units generate a reserved resource status table for each computing power domain, and the short-cycle decision-making units monitor terminal tasks and mark the set of task requests that require cross-domain collaboration. The collaborative control center receives the reserved resource status table and the set of cross-domain collaborative task requests, publishes the heat index and the historical performance rate of each computing power domain at the virtual boundary node, and generates a set of temporary resource trading pairs and associated quotations and credit data for each computing power domain's short-cycle decision-making unit; The long-term decision-making unit of the participating computing power domain receives the set of temporary resource trading pairs and associated price and credit data, compares them with the original reserved resource status table of the domain to calculate the degree of impact, and generates the final cross-domain resource trading agreement and the updated reserved resource status table after negotiation. The short-cycle decision-making unit receives the final cross-domain resource transaction agreement, distributes and executes tasks according to the updated reserved resource status table, records the actual consumption and updates the local reputation data. Each computing power domain reports the actual consumption and the updated local reputation data to the collaborative control center to generate the next round of popularity index and global reputation data.

2. The decoupled computing method based on distributed computing power scheduling as described in claim 1, characterized in that: The construction of the hierarchical computing power domain structure includes the following steps. Obtain the physical location coordinates and network topology connections of each computing power hub, and divide computing power hubs that are physically adjacent and have close network connections into the same computing power domain; Long-cycle decision-making units and short-cycle decision-making units are deployed in each computing power domain, and virtual boundary nodes are set between adjacent computing power domains to represent the resource interaction relationship between adjacent computing power domains.

3. The decoupled computing method based on distributed computing power scheduling as described in claim 2, characterized in that: The long-cycle decision-making unit generates a reserved resource status table for each computing power domain, and the short-cycle decision-making unit monitors terminal tasks and marks the set of task requests requiring cross-domain collaboration, including the following steps. The long-cycle decision-making unit collects historical data on the resource reserves of each computing node in the domain at preset time intervals, predicts future resource demand based on the historical data on resource reserves, and converts it into a reserved resource status table for the domain. The short-cycle decision unit receives terminal task requests in real time, analyzes computing power requirements and task volume, and matches them with the reserved resource status table of this domain. When there are sufficient available resources in this domain and the reserved resources are not occupied, the terminal task request is executed within this domain; otherwise, the terminal task request is marked as a cross-domain collaborative task request and a set of cross-domain collaborative task requests is generated.

4. The decoupled computing method based on distributed computing power scheduling as described in claim 1, characterized in that: The generation of the popularity index and the historical fulfillment rate of each computing power domain includes the following steps: The collaborative control center receives real-time load, idle resource quantity, actual consumption in the previous cycle, and local reputation data reported by each computing power domain; For each virtual boundary node, calculate the resource supply-demand ratio for the real-time load and idle resources of the associated adjacent computing power domains, and determine the heat index based on the resource supply-demand ratio. The historical fulfillment rate of each computing power domain is calculated based on the degree of matching between the actual consumption in the previous cycle and the agreed resource quantity in the corresponding final cross-domain resource transaction agreement.

5. The decoupled computing method based on distributed computing power scheduling as described in claim 4, characterized in that: The generation of the quote includes the following steps: Each computing power domain's short-cycle decision-making unit acquires the heat index, historical fulfillment rate, and real-time load of its domain, and calculates the idle cost of its domain's resources; Expected revenue is calculated based on popularity index, historical fulfillment rate, real-time load of the domain, and resource idle cost; The type and quantity of computing power that can be released are determined based on expected returns, and pricing information is generated by combining the priority weights corresponding to historical performance rates.

6. The decoupled computing method based on distributed computing power scheduling as described in claim 5, characterized in that: The process of generating a set of temporary resource trading pairs and associated quotes and reputation data includes the following steps: Virtual boundary nodes collect bidding information submitted by adjacent computing power domains and parse the type and quantity of releaseable computing power in each bidding information; The available computing power type is matched with the computing power requirements of the task request set that requires cross-domain collaboration, and paired sequentially according to the priority weight in the quotation information to generate temporary resource transaction pairs. Each temporary resource transaction pair includes the initiating domain, the receiving domain, the transaction resource type, and the transaction resource quantity. Each temporary resource trading pair is associated with its corresponding price information and the reputation data of the initiating domain, generating a set of temporary resource trading pairs and associated price and reputation data.

7. The decoupled computing method based on distributed computing power scheduling as described in claim 6, characterized in that: The calculation of the degree of influence includes the following steps: Extract transaction items related to this domain and associated quotation information and reputation data from the temporary resource transaction pair set, and read the original reserved resource status table of this domain. For each transaction, calculate the proportion of the transaction resource quantity to the total resource quantity of the domain during the same period, and determine the impact value based on the proportion; The impact value is compared with the preset fluctuation tolerance range. If the impact value is lower than the lower limit of the tolerance range, it is marked as low impact; if it is between the tolerance range, it is marked as medium impact; and if it is higher than the upper limit of the tolerance range, it is marked as high impact. Based on the marking results, the confirmation status of each transaction in this domain is generated. The confirmation status of low-impact transactions is "confirmed", the confirmation status of high-impact transactions is "needs correction", and the confirmation status of medium-impact transactions is determined based on the comparison between the associated reputation data and the preset reputation threshold.

8. The decoupled computing method based on distributed computing power scheduling as described in claim 7, characterized in that: The process involves generating the final cross-domain resource transaction agreement and the updated reserved resource status table. Includes the following steps, Collect the confirmation status and correction requests submitted by each transaction domain for each transaction item. For transaction items with inconsistent confirmation status, send the correction request of the transaction item to the corresponding transaction domain at the virtual boundary node. Each transaction domain adjusts the quotation information according to the correction request and resubmits the confirmation status until the confirmation status of the transaction items is consistent, thus forming the final cross-domain resource transaction agreement. The long-term decision-making units of each transaction domain adjust the reserved resource status table of their domain according to the number of transaction resources confirmed in the final cross-domain resource transaction agreement, and generate an updated reserved resource status table.

9. The decoupled computing method based on distributed computing power scheduling as described in claim 8, characterized in that: The process of distributing and executing tasks based on the updated reserved resource status table includes the following steps: The short-cycle decision-making unit obtains the final cross-domain resource transaction agreement from the collaborative control center and the updated reserved resource status table from the long-cycle decision-making unit; When a new terminal task request arrives, the computing power requirement of the terminal task request is parsed, and the available resources in this domain are queried in the updated reserved resource status table. When the available resources in this domain are sufficient and the reservation constraints in the updated reserved resource status table are not violated, the terminal task request will be distributed to the computing power nodes in this domain for execution; otherwise, according to the receiving domain determined by the final cross-domain resource transaction protocol, the terminal task request will be forwarded to the virtual boundary node corresponding to the receiving domain for execution. After execution, the execution result and actual consumption are returned to the initiating domain, the actual consumption is recorded, and the local reputation data of this domain is updated according to the performance of the final cross-domain resource transaction agreement.

10. A decoupled computing system based on distributed computing power scheduling, based on the decoupled computing method based on distributed computing power scheduling as described in any one of claims 1 to 9, characterized in that: include, The computing power domain construction module builds a hierarchical structure of computing power domains. The long-cycle decision unit generates a reserved resource status table for each computing power domain, and the short-cycle decision unit monitors terminal tasks and marks the set of task requests that require cross-domain collaboration. The collaborative pricing module receives the reserved resource status table and the set of cross-domain collaborative task requests from the collaborative control center. It publishes the heat index and the historical performance rate of each computing power domain at the virtual boundary node. The short-cycle decision-making units of each computing power domain submit quotations and generate a set of temporary resource trading pairs, as well as associated quotations and credit data. The protocol negotiation module receives a set of temporary resource trading pairs and associated price and credit data from the long-term decision-making unit of the transaction computing power domain. It compares the data with the original reserved resource status table of the domain to calculate the degree of impact. After negotiation, it generates the final cross-domain resource trading agreement and the updated reserved resource status table. The execution feedback module receives the final cross-domain resource transaction agreement, distributes and executes tasks according to the updated reserved resource status table, records the actual consumption and updates the local reputation data. Each computing power domain reports the actual consumption and updated local reputation data to the collaborative control center to generate the next round of popularity index and global reputation data.

Images