A method and device for evaluating load space-time transfer potential of a data center, a terminal device, and a storage medium

Abstract

This invention discloses a method, apparatus, terminal equipment, and storage medium for assessing the spatiotemporal load transfer potential of a data center, relating to the field of load assessment technology. The method includes: acquiring load transfer assessment parameters for the data center; determining a divisibility coefficient based on the maximum number of parallel subtasks and a preset parallelism benchmark value; determining a time flexibility coefficient based on the latest completion deadline and expected execution duration; determining a state dependency coefficient based on the total amount of migration state data and a preset state quantity benchmark value; determining a spatial transferability coefficient based on available link bandwidth, scheduling period length, data volume required for task migration, and data center transmission distance; and weighted summing of the above coefficients to obtain a load spatiotemporal transfer potential coefficient, thereby determining the load spatiotemporal transfer potential. By implementing this invention, the problem of prior art using only maximum tolerable latency to measure flexibility, leading to a deviation between the assessment results and actual schedulable capacity, can be solved, improving the matching degree of potential assessment results.

Description

Technical Field

[0001] This invention relates to the field of load assessment technology, and in particular to a method, apparatus, terminal equipment, and storage medium for assessing the spatiotemporal load transfer potential of a data center. Background Technology

[0002] With the rapid development of cloud computing, big data, and artificial intelligence technologies, the number and scale of data centers continue to grow, highlighting their increasing importance as power loads. At the same time, the integration of a high proportion of renewable energy sources poses serious challenges to power system frequency stability and congestion management. How to leverage the load flexibility of data centers to both reduce their own operating costs and provide regulation services to the power grid has become a research hotspot in both academia and industry.

[0003] Data center load scrambling refers to the flexible transfer of computing tasks in both time (delayed or advanced execution) and spatial dimensions (cross-data center migration) to respond to electricity market price signals or grid regulation needs. This flexibility can be likened to a virtual power transmission service, which alleviates grid congestion by replacing physical power transmission with load scrambling.

[0004] However, existing load spatiotemporal shift potential assessments abstract all computing tasks into black boxes that can be completed within a certain time window, using only the maximum tolerable latency to measure flexibility. This leads to the misjudgment of tasks that are indivisible, strongly state-dependent, and must be executed continuously as having high scheduling flexibility, assuming that they can be arbitrarily time-shifted or time-shifted as long as they are within the latency window. In reality, such tasks are almost impossible to interrupt or migrate. Furthermore, the adjustment boundaries of divisible, stateless, and highly time-elastic tasks are limited to the maximum tolerable latency, ignoring their shardable and distributed execution characteristics. This fails to unleash the true adjustment potential, resulting in a deviation between the load spatiotemporal shift potential assessment results and the actual schedulable capabilities. Summary of the Invention

[0005] This invention provides a method, apparatus, terminal equipment, and storage medium for assessing the potential for load shifting in data centers. It can solve the problem in the prior art that flexibility is measured only by the maximum tolerable latency, which leads to a deviation between the assessment results and the actual schedulable capacity, and improve the matching degree between the potential assessment results and the actual schedulable capacity.

[0006] One embodiment of the present invention provides a method for assessing the spatiotemporal load transfer potential of a data center, comprising: Obtain the load transfer assessment parameters for the data center to be evaluated; the load transfer assessment parameters include: maximum number of parallel subtasks, latest completion time, estimated execution time, total amount of migration status data, available link bandwidth, scheduling period length, amount of data required for task migration, and data center transmission distance; The decomposability coefficient is determined based on the ratio of the maximum number of parallel subtasks to the preset parallelism baseline value. Based on the latest completion deadline and the expected execution time, the task slack is determined, and the task slack is normalized and truncated to obtain the time elasticity coefficient. The state dependency coefficient is determined based on the ratio of the total amount of migration state data to the preset state quantity benchmark value. The spatial transferability coefficient is determined based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance. The load spatiotemporal transfer potential coefficient is obtained by weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient. The load spatiotemporal shift potential of the data center to be evaluated is determined based on the load spatiotemporal shift potential coefficient.

[0007] Furthermore, before obtaining the load spatiotemporal transfer potential coefficient by weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient, the following steps are also included: Obtain the operational objectives of the data center to be evaluated; Based on the operational objectives, query the preset weight factor configuration table to determine the weight factors; The load spatiotemporal transfer potential coefficient is obtained by weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient, including: Based on the weighting factors, the divisibility coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient.

[0008] Furthermore, before obtaining the load spatiotemporal transfer potential coefficient by weighted summation of the descalability coefficient, temporal elasticity coefficient, state dependence coefficient, and spatial transferability coefficient based on weighting factors, the following steps are also included: The task type is determined based on the divisibility coefficient, time elasticity coefficient, and state dependency coefficient. The task types include: overall non-delayable task, overall delayable task, overall state-dependent task, divisible non-delayable task, divisible state-dependent task, or strong locality task. Based on the task type, the divisibility coefficient, temporal elasticity coefficient, state dependency coefficient, and spatial transferability coefficient are modified to obtain the modified divisibility coefficient, modified temporal elasticity coefficient, modified state dependency coefficient, and modified spatial transferability coefficient. Based on weighting factors, the descalability coefficient, temporal resilience coefficient, state dependence coefficient, and spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient, including: Based on the weighting factors, the modified descalability coefficient, modified time elasticity coefficient, modified state dependence coefficient, and modified spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient.

[0009] Furthermore, based on the task type, the descalability coefficient, temporal resilience coefficient, state dependency coefficient, and spatial transferability coefficient are modified to obtain the modified descalability coefficient, modified temporal resilience coefficient, modified state dependency coefficient, and modified spatial transferability coefficient, including: If the task type is a completely non-delayable task or a split non-delayable task, set the time elasticity coefficient to 0 to obtain the corrected time elasticity coefficient; otherwise, use the time elasticity coefficient as the corrected time elasticity coefficient. If the task type is an overall non-delayable task, an overall delayable task, or an overall state-dependent task, the divisibility coefficient is set to 0 to obtain the corrected divisibility coefficient; otherwise, the divisibility coefficient is used as the corrected divisibility coefficient. When the task type is a global state-dependent task, a separable state-dependent task, or a strongly local task, the state dependency coefficient is set to 1 to obtain the corrected state dependency coefficient, and the spatial transferability coefficient is set to 0 to obtain the corrected spatial transferability coefficient; otherwise, the state dependency coefficient is used as the corrected state dependency coefficient, and the spatial transferability coefficient is used as the corrected spatial transferability coefficient.

[0010] Furthermore, based on the divisibility coefficient, time flexibility coefficient, and state dependency coefficient, the task type is determined, including: If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be an overall non-delayable task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be an overall delayable task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, the time elasticity coefficient is less than the second preset time elasticity threshold, and the state dependence coefficient is not less than the preset state dependence threshold, the task type is determined to be an overall state-dependent task; wherein, the first preset time elasticity threshold is less than the second preset time elasticity threshold. If the divisibility coefficient is not less than the preset splitting threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be a divisible but non-delayable task. If the divisibility coefficient is not less than the preset splitting threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, the time elasticity coefficient is less than the second preset time elasticity threshold, and the state dependency coefficient is not less than the preset state dependency threshold, the task type is determined to be a divisible state dependency task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is not less than the preset state dependency threshold, the task type is determined to be a strongly local task.

[0011] Furthermore, based on the latest completion deadline and the estimated execution duration, the task slack is determined, including: Get the current evaluation time in real time; The remaining schedulable time for the task is determined based on the difference between the latest completion deadline and the current assessment time. The sum of the estimated execution time and the preset zero-positive number will be used as the baseline execution time for the task. The task slack is calculated as the ratio of the remaining schedulable time of the task to the baseline execution time of the task.

[0012] Furthermore, based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance, the spatial transferability coefficient is determined, including: Calculate the amount of data that the link can transmit within the scheduling period based on the available bandwidth of the link and the length of the scheduling period; The bandwidth adaptation factor is obtained by normalizing and truncating the ratio of the amount of data that the link can transmit to the amount of data required for task migration. The distance adaptation factor is calculated based on the data center transmission distance and the preset acceptable distance threshold. The spatial transferability coefficient is obtained by weighted summation of the bandwidth adaptation factor and the distance adaptation factor.

[0013] Based on the above method embodiments, the present invention provides corresponding device embodiments, including: an evaluation parameter acquisition module, a descalability calculation module, a time elasticity calculation module, a state dependency calculation module, a spatial transferability calculation module, a transfer potential coefficient calculation module, and a transfer potential evaluation module; The evaluation parameter acquisition module is used to acquire the load transfer evaluation parameters of the data center to be evaluated. The load transfer evaluation parameters include: maximum number of parallel subtasks, latest completion time, estimated execution time, total amount of migration status data, available link bandwidth, scheduling period length, amount of data required for task migration, and data center transmission distance. The splittableness calculation module is used to determine the splittableness coefficient based on the ratio of the maximum number of parallel subtasks to the preset parallelism benchmark value. The time elasticity calculation module is used to determine the task slack based on the latest completion deadline and the expected execution time, and to normalize and truncate the task slack to obtain the time elasticity coefficient. The state dependency calculation module is used to determine the state dependency coefficient based on the ratio of the total amount of migration state data to the preset state quantity benchmark value. The spatial transferability calculation module is used to determine the spatial transferability coefficient based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance. The load transfer potential coefficient calculation module is used to perform a weighted summation of the decomposability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient to obtain the load spatiotemporal transfer potential coefficient. The load transfer potential assessment module is used to determine the load transfer potential of the data center to be assessed based on the load transfer potential coefficient.

[0014] Based on the above method embodiments, the present invention provides a corresponding terminal device embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps of the data center load spatiotemporal transfer potential assessment method as described in the present invention.

[0015] Based on the above method embodiments, the present invention provides a corresponding computer-readable storage medium embodiment, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the data center load spatiotemporal transfer potential assessment method as described in the present invention.

[0016] Compared with the prior art, the beneficial effects of this embodiment are as follows: This invention acquires load transfer assessment parameters for the data center to be evaluated, including the maximum number of parallel subtasks, the latest completion deadline, the estimated execution time, the total amount of migration status data, the available bandwidth of the link, the scheduling period length, the amount of data required for task migration, and the data center transmission distance. Based on the ratio of the maximum number of parallel subtasks to a preset parallelism benchmark, a splittable coefficient is determined, which directly determines whether a task can be split in parallel. The lower the splittable coefficient, the more difficult the task is to split and the stronger its rigidity, thus effectively eliminating misjudgments of indivisible rigid tasks. Based on the latest completion deadline and the estimated execution time, the task slack is determined, and the task slack is normalized and truncated to obtain a time elasticity coefficient. The higher the time elasticity coefficient, the more sufficient the time margin and the stronger the scheduling elasticity, accurately characterizing the time adjustability potential of different tasks. Based on the ratio of the total amount of migration status data to a preset state quantity benchmark, a state dependency coefficient is determined, quantifying the degree of task state binding. The lower the state dependency coefficient, the stronger the state dependency and the greater the migration overhead, thus effectively identifying strongly dependent tasks and avoiding their misjudgment as schedulable load. Based on available link bandwidth, scheduling period length, data volume required for task migration, and data center transmission distance, a spatial transferability coefficient is determined to quantify the feasibility of task migration across data centers. A higher spatial transferability coefficient indicates stronger spatial transferability, filling the gap in existing spatial dimension assessment technologies and avoiding the disconnect between assessment and reality caused by spatial constraints. A weighted sum of the descalability coefficient, temporal elasticity coefficient, state dependency coefficient, and spatial transferability coefficient comprehensively considers the task's descalability, temporal elasticity, state dependency, and spatial transfer feasibility to obtain a load spatiotemporal transfer potential coefficient. This coefficient determines the load spatiotemporal transfer potential of the data center to be evaluated, providing a comprehensive and realistic characterization of the task's actual schedulable capability.

[0017] In summary, this invention collects and quantifies multi-dimensional parameters, including the divisibility coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient. This allows the assessment of load scheduling capability to go beyond a single time delay indicator, comprehensively reflecting task splitting characteristics, time margin, state transition overhead, and spatial transmission feasibility. This solves the problem in existing technologies where flexibility is measured solely by the maximum tolerable delay, leading to a deviation between the assessment results and actual schedulable capability. It also improves the matching degree between potential assessment results and actual schedulable capability. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for assessing the spatiotemporal load transfer potential of a data center according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a data center load spatiotemporal transfer potential assessment device provided in an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated.

[0021] like Figure 1 As shown, in order to address the problem that existing technologies measure flexibility solely by maximum tolerable latency, leading to discrepancies between assessment results and actual schedulable capabilities, an embodiment of the present invention provides a method for assessing the spatiotemporal load transfer potential of a data center. This method includes at least the following steps: Step S1: Obtain the load transfer assessment parameters for the data center to be evaluated; the load transfer assessment parameters include: maximum number of parallel subtasks, latest completion time, estimated execution time, total amount of migration status data, available link bandwidth, scheduling period length, amount of data required for task migration, and data center transmission distance; For step S1, load transfer evaluation parameters corresponding to all computing power tasks within the data center to be evaluated are obtained from the database. The load transfer evaluation parameters for each computing power task include: task... Maximum number of parallel subtasks that can be supported under system-allowed conditions ,Task Latest completion deadline ,Task Expected execution time ,Task The total amount of state data of the memory image, disk volume, or database to be transferred. (i.e., total migration status data, unit: GB), Task Available bandwidth of the link ,Task Scheduling period length ,Task Data volume required for migration and the physical distance between the source data center and the target data center (i.e., data center transmission distance, unit: km), usually taken as the straight-line distance or fiber optic route distance between the two locations.

[0022] It should be noted that the computing power tasks mentioned in this invention can be uniformly referred to as "tasks" in the following sections, as they are the same expression.

[0023] Step S2: Determine the decomposability coefficient based on the ratio of the maximum number of parallel subtasks to the preset parallelism baseline value; For step S2, for each computing task, based on the maximum number of parallel subtasks... With preset parallelism benchmark value The ratio of the divisibility coefficient is calculated using the following formula: ; in, Indicates the first The decomposability coefficient of a task is used to characterize whether a computing task can be decomposed into multiple independently executable subtasks, as well as the maximum granularity at which it can be decomposed in parallel. This represents a truncation function used to truncate... Limited to the range This is done to normalize the numerical values ​​and avoid the impact of extreme values ​​on subsequent scheduling calculations. Indicates the first The maximum number of parallel subtasks for a task is a direct reflection of the task's parallel processing capability. This represents the preset parallelism baseline value, which can be determined based on the typical maximum concurrency allowed by the data center platform or the high quantile of the historical task concurrency distribution.

[0024] Based on the above calculation logic, the value of the divisibility coefficient directly corresponds to the parallel scheduling potential of the task. The closer the divisibility coefficient is to 1, the stronger the parallel splitting potential of the task, which can be split into more independent subtasks to achieve distributed parallel execution. This can make full use of the distributed computing resources of multiple data centers and improve task execution efficiency and data center resource utilization. Conversely, the closer the divisibility coefficient is to 0, the weaker the parallelism of the task, making it difficult to split and schedule, and it is only suitable for execution as a complete task.

[0025] Preferably, a preset splitting threshold is set. In this embodiment, a preset splitting threshold is used. .

[0026] By setting a splitting threshold Discretize the separability into two states, specifically, when the... When the divisibility coefficient of a task is less than a preset splitting threshold, that is... Determine the first The task is an indivisible whole; when the first task is... When the divisibility coefficient of a task is greater than or equal to a preset splitting threshold, that is... Determine the first Each task is divisible.

[0027] Step S3: Determine the task slack based on the latest completion deadline and the expected execution time, and normalize and truncate the task slack to obtain the time elasticity coefficient; In a preferred embodiment, determining task slack based on the latest completion deadline and the expected execution duration includes: Get the current evaluation time in real time; The remaining schedulable time for the task is determined based on the difference between the latest completion deadline and the current assessment time. The sum of the estimated execution time and the preset zero-positive number will be used as the baseline execution time for the task. The task slack is calculated as the ratio of the remaining schedulable time of the task to the baseline execution time of the task.

[0028] For step S3, the current evaluation time is obtained in real time. For each computing task, based on the latest completion deadline Compared to the current assessment time The difference is used to calculate the remaining schedulable time of the task. ; Expected execution time With preset protection against zero positive numbers The sum is used as the baseline execution time for the task. Meanwhile, the task slack is calculated by the ratio of the remaining schedulable time of the task to the baseline execution time of the task. The specific calculation formula is as follows: ; in, Indicates the first Task slack for each task Indicates the first The latest completion time for each task Indicates the current assessment time. Indicates the first The estimated execution time of each task. This indicates a preset protection against removing positive zeros, used to avoid exceeding the estimated execution time. Division by zero exception when the value is zero.

[0029] By calculating task slackness, the time urgency and scheduling flexibility of a task are quantitatively represented. The larger the task slackness value, the more abundant the remaining schedulable time of the task relative to the baseline execution time, the more relaxed the time constraint of the task, and the higher the scheduling flexibility. Time-consuming operations such as cross-data center migration and splitting can be prioritized. The smaller the task slackness value, the tighter the remaining schedulable time of the task, the stronger the time constraint. Local execution should be prioritized to avoid the latency risk caused by migration and ensure that the task is completed on time.

[0030] Next, the task relaxation is normalized and truncated using a truncation function to obtain the time elasticity coefficient, as shown in the following formula: ; in, Indicates the first The time elasticity coefficient of each task is used to characterize the urgency of the computing power task in terms of execution completion time; and This indicates the preset relaxation baseline, which is determined by the low and high quantiles of historical relaxation. When there is no historical data, the minimum and maximum relaxation values ​​that the system can accept can be taken from engineering experience.

[0031] Preferably, a first preset time elasticity threshold is set. Second preset time elastic threshold Among them, the first preset time elasticity threshold Less than the second preset time elasticity threshold In this embodiment, the first preset time elastic threshold Second preset time elastic threshold The quantiles are determined by the distribution of the time elasticity coefficient of historical tasks, taking the 10th percentile and the 70th percentile.

[0032] By the first preset time elastic threshold Second preset time elastic threshold The time elasticity is divided into three states. Specifically, when the... When the time elasticity coefficient of a task is less than the first preset time elasticity threshold, that is... Determine the first The first task is inflexible; when the first task is inflexible... The time elasticity coefficient of the first task is not less than the first preset time elasticity threshold, and the time elasticity coefficient of the second task is not less than the first preset time elasticity threshold. When the time elasticity coefficient of a task is less than the second preset time elasticity threshold, that is... Determine the first The first task is weakly elastic; when the first task is weakly elastic. When the time elasticity coefficient of each task is not less than the second preset time elasticity threshold, that is... Determine the first Each task is highly flexible.

[0033] Step S4: Determine the state dependency coefficient based on the ratio of the total amount of migration state data to the preset state quantity benchmark value; For step S4, for each computing task, based on the total amount of migration state data... Compared with the preset state quantity reference value The ratio of is used to calculate the state dependence coefficient using the following formula: ; in, Indicates the first The state dependency coefficient of a task is used to characterize the degree to which a computing task is bound to memory data, storage state, or specific hardware resources during execution. Indicates the first The total amount of memory images, disk volumes, or database status data that need to be transferred for each task transfer; This represents the preset state quantity reference value.

[0034] Based on the above calculation logic, the magnitude of the state dependency coefficient directly corresponds to the degree to which the task is bound to the original operating environment, and the cost and feasibility of cross-data center migration. The closer the state dependency coefficient is to 1, the larger the total amount of state data such as memory images, disk volumes, or databases that need to be transferred during task migration. The higher the degree of binding to the operating environment, storage resources, and hardware status of the source data center, the greater the network transmission overhead required for cross-data center migration, the longer the migration latency, the higher the risk of migration failure, and the lower the feasibility of task migration. It is more suitable to execute the task locally in the source data center. Conversely, the closer the state dependency coefficient is to 0, the smaller the scale of the task's state data, the lower the dependence on the original operating environment, the lower the network transmission cost and complexity during migration, the smaller the migration time and risk, and the higher the feasibility of task migration. It is more suitable to achieve distributed parallel execution through cross-data center load balancing to improve the utilization rate of global computing resources.

[0035] Preferably, a first preset state discrimination threshold is set. Second preset state discrimination threshold Among them, the first preset state discrimination threshold Less than the second preset state discrimination threshold In this embodiment, the first preset state discrimination threshold Second preset state discrimination threshold The coefficients are adaptively determined based on the statistical distribution of historical task state dependency coefficients.

[0036] The threshold is determined by the first preset state. Second preset state discrimination threshold State dependencies are categorized into three types of states. Specifically, when the first... When the state dependency coefficient of a task is less than the first preset state discrimination threshold, that is... Determine the first The task is weakly state dependent; when the task is... The state dependency coefficient of each task is not less than the first preset state discrimination threshold, and the first task's state dependency coefficient is not less than the first preset state discrimination threshold. When the state dependency coefficient of a task is less than the second preset state discrimination threshold, that is... Determine the first The task is state-dependent; when the first task is state-dependent; When the state dependency coefficient of each task is not less than the second preset state discrimination threshold, that is... Determine the first Each task has a strong state dependency.

[0037] It should be noted that the present invention uses the second preset state discrimination threshold. This serves as a preset state dependency threshold, allowing tasks with weak and medium state dependencies to participate in spatial transfer.

[0038] Step S5: Determine the spatial transferability coefficient based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance; In a preferred embodiment, the spatial transferability coefficient is determined based on the available link bandwidth, scheduling period length, data volume required for task migration, and data center transmission distance, including: Calculate the amount of data that the link can transmit within the scheduling period based on the available bandwidth of the link and the length of the scheduling period; The bandwidth adaptation factor is obtained by normalizing and truncating the ratio of the amount of data that the link can transmit to the amount of data required for task migration. The distance adaptation factor is calculated based on the data center transmission distance and the preset acceptable distance threshold. The spatial transferability coefficient is obtained by weighted summation of the bandwidth adaptation factor and the distance adaptation factor.

[0039] For step S5, for each computing task, based on the available bandwidth of the link... With the length of the scheduling period The product of the two factors is used to calculate the amount of data that the link can transmit during the scheduling period, thus characterizing the data transmission capacity that the network link can carry within a unit scheduling cycle.

[0040] The amount of data that the link can transmit during the scheduling period and the amount of data required for task migration are compared. The ratio is calculated, and then normalized and truncated using a truncation function to obtain the bandwidth adaptation factor. The specific formula is as follows: ; in, Indicates the first The bandwidth adaptation factor for each task is used to quantify the degree of matching between the link transmission capacity and the task's data migration requirements. If the link transmission capacity is sufficient to cover the amount of data to be migrated, the bandwidth adaptation factor remains within a reasonable range; if the capacity is insufficient, the bandwidth adaptation factor is limited to the boundary value. Indicates the first The available bandwidth of the link for each task; Indicates the length of the scheduling period; Indicates the first The amount of data required for task migration for each task.

[0041] Next, based on the data center transmission distance With respect to the preset acceptable distance threshold The ratio of the distance fit factor is calculated using the following formula: ; in, Indicates the first The distance adaptation factor for each task is used to quantify the impact of the physical distance between the source data center and the target data center on the feasibility of cross-domain migration of the task. Indicates the first The data center transmission distance for each task is used to characterize the physical distance between the source data center and the target data center of the task. This indicates a preset acceptable distance threshold, used to define the engineering feasibility boundary for cross-data center migration.

[0042] It should be noted that the source data center refers to the data center to which the task to be transferred belongs; the target data center refers to the data center to which the task belongs after the transfer.

[0043] Next, the bandwidth adaptation factor With distance adaptation factor We perform a weighted summation to obtain the spatial transferability coefficients, as shown in the following formula: ; in, Indicates the first The spatial transferability coefficient of each task is used to quantitatively evaluate the feasibility of migrating computing power tasks across data centers. The closer the spatial transferability coefficient is to 1, the stronger the migration adaptability and the lower the cost, and cross-domain migration should be prioritized. The closer the spatial transferability coefficient is to 0, the worse the migration feasibility, and local execution should be prioritized. and These represent the weights of the bandwidth adaptation factor and the distance adaptation factor, respectively. In this embodiment, the distance factor comprehensively considers transmission latency and cross-domain costs, and its importance is greater than the bandwidth-to-data-volume ratio. Therefore, it is set... , Further optimizations will be made based on the data center's operational status.

[0044] Step S6: Perform a weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient to obtain the load spatiotemporal transfer potential coefficient; For step S6, the divisibility coefficient Time elasticity coefficient State dependency coefficient and spatial transferability coefficient The load spatiotemporal transfer potential coefficient is calculated by performing a weighted summation, as shown in the following formula: ; in, Indicates the first The load spatiotemporal transfer potential coefficient of each task. The value range is [0,1]. The closer the value is to 1, the higher the potential for spatiotemporal transfer of computing power tasks; The closer the value is to 0, the lower the potential for spatiotemporal transfer of computing power tasks; , , and The weighting coefficient, also called the weighting factor, satisfies... In this invention, the weighting coefficient , , and (i.e., weighting factor), used to balance the contribution ratio of time elasticity coefficient, spatial transferability coefficient, decomposability coefficient and state dependence coefficient in the calculation of load spatiotemporal transfer potential.

[0045] Preferably, the weighting factor can be preset based on industry experience, typical data center operating scenarios, or historical scheduling data.

[0046] To adapt to the differentiated operational needs of different data centers, this invention also supports dynamically determining weighting factors based on operational objectives. The specific process is as follows: In a preferred embodiment, before obtaining the load spatiotemporal transfer potential coefficient by weighted summation of the descalability coefficient, temporal elasticity coefficient, state dependence coefficient, and spatial transferability coefficient, the method further includes: Obtain the operational objectives of the data center to be evaluated; Based on the operational objectives, query the preset weight factor configuration table to determine the weight factors; The load spatiotemporal transfer potential coefficient is obtained by weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient, including: Based on the weighting factors, the divisibility coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient.

[0047] In one embodiment of the present invention, the operational objectives of the data center to be evaluated are obtained, wherein the operational objectives include minimizing operating costs, minimizing carbon emissions, maximizing Service Level Agreement (SLA) compliance rate, and balancing and optimizing or maximizing the revenue of virtual transmission services.

[0048] Subsequently, based on the operational objectives, the preset weight factor configuration table (Table 1) is queried to match the weight factor values ​​that are suitable for the current operational objectives. For example, when the operational objective is to minimize operating costs, the weight corresponding to the space transferability coefficient is increased. This strengthens the influence of spatial dimensions on load shifting decisions, prioritizing the spatial shift of tasks to target data centers with lower electricity prices and better bandwidth costs; when the operational objective is to minimize carbon emissions, the weight corresponding to the time elasticity coefficient is increased. Carbon emission reduction is achieved by adapting to periods of abundant green electricity through time-delay scheduling.

[0049] Table 1. Weighting Factor Configuration under Different Operational Objectives Finally, based on the weighting factors determined by matching, the divisibility coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient.

[0050] Through the aforementioned weight configuration, this invention achieves personalized and dynamic weight adaptation for different operational objectives, significantly improving the scenario adaptability and flexibility of load transfer scheduling strategies, and ensuring that the calculation of the load spatiotemporal transfer potential coefficient always aligns with the actual operational needs of the data center.

[0051] Taking the same computing power task as an example, the complete calculation process of its load spatiotemporal transfer potential coefficient is explained: First, calculate the divisibility coefficient. This task can be trained on multiple GPUs using a data-parallel approach, with a maximum number of parallel subtasks. This equals the number of available GPUs. If the number of available GPUs in the data center is 32, the default parallelism baseline value is... If we set it to 64, then the task's divisibility coefficient is... If a preset splitting threshold is used... If we take 0.2, then 0.5 ≥ 0.2, and the task is determined to be a divisible task.

[0052] Next, the time elasticity coefficient is calculated, assuming the current evaluation time. And at the current moment There are 10ms left until the latest completion deadline for the task. Estimated execution time: If it is 5ms, then the task slack is... If a relaxation baseline is preset , Then the time flexibility coefficient of the task If the first preset time elasticity threshold The second preset time elastic threshold If 0.1≤0.6≤0.7, then the task is determined to be a weakly elastic task.

[0053] Subsequently, the state dependency coefficient is calculated. This task is a stateful database task, and its memory usage is... Take 80GB, preset state quantity reference value Given 100GB, the state dependency coefficient of this task is... If the first preset state discrimination threshold is... Second preset state discrimination threshold If 0.8 ≥ 0.7, then the task is determined to be a strongly state-dependent task. .

[0054] Next, the spatial transferability coefficient is calculated, assuming the available bandwidth of the link between the source data center and the target data center. The speed is 10Gbps, and the scheduling period length is... Taking 1 hour (i.e., 3600 seconds) as an example, the amount of data required for task migration. If the bandwidth is 100GB, then the bandwidth adaptation factor for this task is... If the data center transmission distance Preset acceptable distance threshold Then the distance fit factor for this task If the bandwidth adaptation factor weights Distance adaptation factor weight Then the spatial transferability coefficient of the task .

[0055] It should be noted that the unit conversion is as follows: 10Gbps = 10×10^9 bps, and the amount of data that can be transmitted in 1 hour is 10×10^9×3600 bits = 3.6×10^13 bits = 4.5×10^12 bytes≈4.5 TB, which is much larger than 100GB.

[0056] Based on the above coefficient values , , , If the data center's operational objective is to maximize revenue from virtual power transmission services, then according to Table 1... , , , Then the spatiotemporal transfer potential coefficient of this task is This value indicates that the task has a moderate to high potential for spatiotemporal transfer and can be considered for inclusion in the application pool for virtual power transmission services.

[0057] It should be noted that in the above scheduling, the non-zero coefficients of weakly characterized tasks may cause the scheduling system to misjudge their transfer potential, resulting in a large number of invalid migrations. To further improve the calculation accuracy of the load spatiotemporal transfer potential coefficient, achieve differentiated scheduling adaptation, and address the problem that traditional coefficient calculations still retain non-zero coefficients for weakly characterized tasks, leading to scheduling decisions deviating from actual task requirements, this invention introduces task type determination and coefficient correction before weighted summation of the descalability coefficient, temporal elasticity coefficient, state dependency coefficient, and spatial transferability coefficient. The specific process is as follows: In a preferred embodiment, before obtaining the load spatiotemporal transfer potential coefficient by weighted summation of the descalability coefficient, temporal elasticity coefficient, state dependence coefficient, and spatial transferability coefficient based on weighting factors, the method further includes: The task type is determined based on the divisibility coefficient, time elasticity coefficient, and state dependency coefficient. The task types include: overall non-delayable task, overall delayable task, overall state-dependent task, divisible non-delayable task, divisible state-dependent task, or strong locality task. Based on the task type, the divisibility coefficient, temporal elasticity coefficient, state dependency coefficient, and spatial transferability coefficient are modified to obtain the modified divisibility coefficient, modified temporal elasticity coefficient, modified state dependency coefficient, and modified spatial transferability coefficient. Based on weighting factors, the descalability coefficient, temporal resilience coefficient, state dependence coefficient, and spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient, including: Based on the weighting factors, the modified descalability coefficient, modified time elasticity coefficient, modified state dependence coefficient, and modified spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient.

[0058] In one embodiment of the present invention, firstly, the task type is determined based on the divisibility coefficient, time elasticity coefficient, and state dependency coefficient, and the computing power tasks are divided into seven categories, including overall non-delayable tasks, overall delayable tasks, overall state-dependent tasks, divisible non-delayable tasks, divisible delayable tasks, divisible state-dependent tasks, or strongly local tasks.

[0059] Preferably, the task type is determined based on the divisibility coefficient, time flexibility coefficient, and state dependency coefficient, including: If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be an overall non-delayable task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be an overall delayable task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, the time elasticity coefficient is less than the second preset time elasticity threshold, and the state dependence coefficient is not less than the preset state dependence threshold, the task type is determined to be an overall state-dependent task; wherein, the first preset time elasticity threshold is less than the second preset time elasticity threshold. If the divisibility coefficient is not less than the preset splitting threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be a divisible but non-delayable task. If the divisibility coefficient is not less than the preset splitting threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, the time elasticity coefficient is less than the second preset time elasticity threshold, and the state dependency coefficient is not less than the preset state dependency threshold, the task type is determined to be a divisible state dependency task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is not less than the preset state dependency threshold, the task type is determined to be a strongly local task.

[0060] Specifically, as shown in Table 2, when the divisibility coefficient Less than the preset splitting threshold Time elasticity coefficient Less than the first preset time elasticity threshold And the state dependency coefficient Less than the preset state dependency threshold When the task type is determined to be an overall non-delayable task, the type code is [type code missing]. These types of tasks do not have the ability to be split for execution, have extremely strong time constraints, and have low dependence on the state of the operating environment. They belong to business tasks with strict latency requirements, such as high-frequency trading and industrial real-time control systems. When the separability coefficient Less than the preset splitting threshold Time elasticity coefficient Not less than the first preset time elasticity threshold And the state dependency coefficient Less than the preset state dependency threshold When the task type is determined to be an overall time-delayable task, the type code is [type code missing]. These types of tasks do not have the ability to be split for execution, have loose time constraints, and have low dependence on the state of the operating environment. They belong to non-real-time business tasks that can be flexibly scheduled, such as offline scientific computing and batch processing of monolithic applications. When the separability coefficient Less than the preset splitting threshold Time elasticity coefficient Not less than the first preset time elasticity threshold Time elasticity coefficient Less than the second preset time elasticity threshold And the state dependency coefficient Not less than the preset state dependency threshold When the task type is determined to be an overall state-dependent task, the type code is: These types of tasks do not have the ability to be split for execution, have a medium time elasticity, and are highly dependent on the state of the runtime environment. They are holistic business tasks that are strongly bound to the local runtime environment, such as state databases and in-memory computing clusters. When the separability coefficient Not less than the preset splitting threshold Time elasticity coefficient Less than the first preset time elasticity threshold And the state dependency coefficient Less than the preset state dependency threshold When the task type is determined to be a splittable, non-delayable task, the type code is [type code missing]. These tasks have the ability to be executed in parallel, have strict time constraints and low dependence on the state of the operating environment. They are business tasks that are sensitive to execution latency but can improve processing speed through distributed parallelism, such as real-time video stream analysis and online AI inference. When the separability coefficient Not less than the preset splitting threshold Time elasticity coefficient Not less than the second preset time elasticity threshold And the state dependency coefficient Less than the preset state dependency threshold When the task type is determined to be a splittable and delayable task, the type code is: These tasks have good parallel splitting and execution capabilities, sufficient time scheduling flexibility and low dependence on the state of the running environment. They are offline computing tasks that are not sensitive to the execution time period and are suitable for large-scale distributed scheduling, such as large-scale AI training and big data offline analysis. When the separability coefficient Not less than the preset splitting threshold Time elasticity coefficient Not less than the first preset time elasticity threshold Time elasticity coefficient Less than the second preset time elasticity threshold And the state dependency coefficient Not less than the preset state dependency threshold When the task type is determined to be a shardable state-dependent task, the type code is: These tasks have the ability to be split and executed in parallel, have a medium time elasticity, and are highly dependent on the state of the operating environment. They are business tasks that need to be executed in parallel in collaboration on the local node or within the near-source data center, such as distributed graph computing and iterative machine learning. When the separability coefficient Less than the preset splitting threshold Time elasticity coefficient Less than the first preset time elasticity threshold And the state dependency coefficient Not less than the preset state dependency threshold When the task type is determined to be a strongly local task, the type code is: Such tasks lack the ability to be split for execution, have strict time constraints, and are highly bound to local hardware environment, physical devices, or data compliance requirements. They belong to strongly bound businesses that must be executed in place at a designated node, such as businesses that rely on dedicated encryption cards, physical signal acquisition equipment, and data sovereignty compliance.

[0061] By combining the divisibility coefficient, time elasticity coefficient, and state dependency coefficient with corresponding preset thresholds in a multi-dimensional combination, various computing power tasks can be accurately classified into seven types: overall non-delayable tasks, overall delayable tasks, overall state-dependent tasks, divisible non-delayable tasks, divisible delayable tasks, divisible state-dependent tasks, and strongly local tasks. This provides a basis for subsequent differential coefficient correction for different task types.

[0062] Table 2. Threshold Comparison Table for Determining Computing Power Task Type Preferably, based on the task type, the descalability coefficient, temporal flexibility coefficient, state dependency coefficient, and spatial transferability coefficient are modified to obtain modified descalability coefficient, modified temporal flexibility coefficient, modified state dependency coefficient, and modified spatial transferability coefficient, including: If the task type is a completely non-delayable task or a split non-delayable task, set the time elasticity coefficient to 0 to obtain the corrected time elasticity coefficient; otherwise, use the time elasticity coefficient as the corrected time elasticity coefficient. If the task type is an overall non-delayable task, an overall delayable task, or an overall state-dependent task, the divisibility coefficient is set to 0 to obtain the corrected divisibility coefficient; otherwise, the divisibility coefficient is used as the corrected divisibility coefficient. When the task type is a global state-dependent task, a separable state-dependent task, or a strongly local task, the state dependency coefficient is set to 1 to obtain the corrected state dependency coefficient, and the spatial transferability coefficient is set to 0 to obtain the corrected spatial transferability coefficient; otherwise, the state dependency coefficient is used as the corrected state dependency coefficient, and the spatial transferability coefficient is used as the corrected spatial transferability coefficient.

[0063] Specifically, after determining the type of computing task, targeted adjustments are made to the divisibility coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient based on the task type to avoid extremely small coefficient values ​​close to 0, which could have a slight impact on the subsequent weighted summation and lead to misjudgment of the task's potential for time, space, or divisibility transfer.

[0064] Specifically, for overall non-delayable tasks Or a split, non-delayable task The time elasticity coefficient is forcibly set to 0, completely blocking the time dimension transfer permission of this type of task; for other task types, the original time elasticity coefficient is retained as the modified time elasticity coefficient, so as to set rigid time constraints for non-delayable tasks, while retaining scheduling flexibility for tasks with time elasticity.

[0065] For overall non-delayed tasks Overall Delayable Tasks Or overall state depends on the task The divisibility coefficient is forcibly set to 0, thereby prohibiting the splitting and scheduling of this type of holistic task from the root. For other task types, only the original divisibility coefficient is retained as the modified divisibility coefficient, thereby imposing an indivisible rigid constraint on holistic tasks, while reserving space for parallel scheduling and distributed execution for tasks that have the conditions for splitting.

[0066] For tasks that depend on the overall state Separable state-dependent tasks or highly localized tasks The state dependency coefficient is forced to 1 to strengthen state constraints, while the spatial transferability coefficient is forced to 0 to completely block cross-data center spatial migration. For other task types, only the original state dependency coefficient and spatial transferability coefficient are retained as the corresponding coefficients after correction. This is to impose strict local execution restrictions on tasks with high state dependency and strong local binding, while retaining the flexibility of cross-node scheduling for migration-friendly tasks.

[0067] Taking a specific computing task as an example, this task is identified as an overall non-delayable task after evaluation. During the coefficient correction process, the time elasticity coefficient and the decomposability coefficient of the task are set to 0, while the state dependence coefficient and the spatial transferability coefficient remain unchanged.

[0068] By correcting the above coefficients, this invention can limit time delay, task splitting, cross-data center migration, etc., according to the execution characteristics of each task, thereby eliminating the interference and misleading effect of weak coefficient values ​​on subsequent potential calculations, and making the subsequently calculated load spatiotemporal transfer potential more closely match the actual schedulable range.

[0069] Based on this, the modified descalability coefficient, modified time elasticity coefficient, modified state dependence coefficient, and modified spatial transferability coefficient are weighted and summed according to the weighting factor to obtain the load spatiotemporal transferability potential coefficient. This allows the final potential coefficient to intuitively and accurately measure the comprehensive transferability of the task in terms of time, space, descalability, and state.

[0070] Step S7: Determine the load spatiotemporal transfer potential of the data center to be evaluated based on the load spatiotemporal transfer potential coefficient.

[0071] For step S7, the load spatiotemporal transfer potential coefficient calculated in step S6 is used. This directly quantifies and assesses the load spatiotemporal shift potential of the data center under evaluation, resulting in a load spatiotemporal shift potential coefficient. A higher value indicates that the task has greater adjustability in terms of time delay, spatial migration, and split execution, and the corresponding stronger load spatiotemporal transfer capability, making it more suitable for participating in resource optimization operations such as cross-node scheduling and off-peak execution; Load spatiotemporal transfer potential coefficient The lower the value, the more rigid constraints the task is subject to, the smaller the schedulable space, and the weaker the potential for load transfer in time and space. It should be executed locally in place first.

[0072] Preferably, capacity assessment of the virtual power transmission service of the data center is performed based on the load spatiotemporal transfer potential coefficient, and the specific process is as follows: First, set a threshold for transfer potential. The load transfer potential coefficient for each computing task in time and space. Respectively with the transfer potential threshold By comparison, the potential coefficient for load transfer in time and space is selected. Greater than the transfer potential threshold All computing power tasks are processed to obtain a set of computing power tasks, thereby eliminating low-potential tasks with weak spatiotemporal adjustment capabilities and no cross-node scheduling value.

[0073] Subsequently, the rated power of each computing task in the computing task set is obtained. Combined with the load spatiotemporal transfer potential coefficient and status indicator variables The data center's time period can be calculated using the following formula. Scheduled capacity: ; in, Indicates the data center during the time period Scheduled capacity, Indicates the first The load spatiotemporal transfer potential coefficient of each task. Indicates the first The rated power of each task, State indicator variables are used to characterize computing tasks. During the period Is the internal system in a schedulable state? This is the index of the computing task in the computing task set.

[0074] It should be noted that the state indicator variable For 0-1 variables, computing power task During the period When the internal system is in a schedulable state, A value of 1 indicates a computing power task. During the period When the internal system is in an unschedulable state, The value is 0.

[0075] By weighted summing of the scheduling status, rated power, and transfer potential of all tasks within the set during the corresponding time period, the aggregate calculation of transferable load is completed, resulting in the dispatchable load curve of the data center for each time period. Using the dispatchable load curve as the virtual power transmission service capacity that the data center can provide, it intuitively reflects the power range that the data center can flexibly adjust in both time and space dimensions.

[0076] like Figure 2 As shown, based on the above method embodiments, corresponding apparatus embodiments are provided; One embodiment of the present invention provides a data center load spatiotemporal transfer potential assessment device, including: an assessment parameter acquisition module, a shardability calculation module, a time elasticity calculation module, a state dependency calculation module, a spatial transferability calculation module, a transfer potential coefficient calculation module, and a transfer potential assessment module; The evaluation parameter acquisition module is used to acquire the load transfer evaluation parameters of the data center to be evaluated. The load transfer evaluation parameters include: maximum number of parallel subtasks, latest completion time, estimated execution time, total amount of migration status data, available link bandwidth, scheduling period length, amount of data required for task migration, and data center transmission distance. The splittableness calculation module is used to determine the splittableness coefficient based on the ratio of the maximum number of parallel subtasks to the preset parallelism benchmark value. The time elasticity calculation module is used to determine the task slack based on the latest completion deadline and the expected execution time, and to normalize and truncate the task slack to obtain the time elasticity coefficient. The state dependency calculation module is used to determine the state dependency coefficient based on the ratio of the total amount of migration state data to the preset state quantity benchmark value. The spatial transferability calculation module is used to determine the spatial transferability coefficient based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance. The load transfer potential coefficient calculation module is used to perform a weighted summation of the decomposability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient to obtain the load spatiotemporal transfer potential coefficient. The load transfer potential assessment module is used to determine the load transfer potential of the data center to be assessed based on the load transfer potential coefficient.

[0077] It is understood that the above-described device embodiments correspond to the method embodiments of the present invention, and can implement the data center load spatiotemporal transfer potential assessment method provided by any of the above-described method embodiments of the present invention.

[0078] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0079] Based on the above-described embodiments of the data center load spatiotemporal shift potential assessment method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the data center load spatiotemporal shift potential assessment method of any embodiment of the present invention.

[0080] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0081] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0082] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting various parts of the terminal device via various interfaces and lines.

[0083] Based on the above-described method embodiments, another embodiment is provided: another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the data center load spatiotemporal transfer potential assessment method described in any of the above-described method embodiments of the present invention.

[0084] The modules / units integrated into the load transfer potential assessment device / terminal equipment of the data center, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0085] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for assessing the potential for load spatiotemporal shifting in a data center, characterized in that, include: Obtain load transfer assessment parameters for the data center to be evaluated; The load transfer assessment parameters include: maximum number of parallel subtasks, latest completion time, estimated execution time, total amount of migration status data, available link bandwidth, scheduling period length, amount of data required for task migration, and data center transmission distance. The decomposability coefficient is determined based on the ratio of the maximum number of parallel subtasks to the preset parallelism baseline value. Based on the latest completion deadline and the expected execution time, the task slack is determined, and the task slack is normalized and truncated to obtain the time elasticity coefficient. The state dependency coefficient is determined based on the ratio of the total amount of migration state data to the preset state quantity benchmark value. The spatial transferability coefficient is determined based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance. The load spatiotemporal transfer potential coefficient is obtained by weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient. The load spatiotemporal shift potential of the data center to be evaluated is determined based on the load spatiotemporal shift potential coefficient.

2. The method for assessing the spatiotemporal load transfer potential of a data center according to claim 1, characterized in that, Before obtaining the load spatiotemporal transfer potential coefficient by weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient, the following steps are also included: Obtain the operational objectives of the data center to be evaluated; Based on the operational objectives, query the preset weight factor configuration table to determine the weight factors; The weighted summation of the descalability coefficient, temporal elasticity coefficient, state dependence coefficient, and spatial transferability coefficient yields the load spatiotemporal transfer potential coefficient, including: Based on the weighting factors, the divisibility coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient.

3. The method for assessing the spatiotemporal load transfer potential of a data center according to claim 2, characterized in that, Before obtaining the load spatiotemporal transfer potential coefficient by weighted summation of the descalability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient based on weighting factors, the following steps are also included: The task type is determined based on the divisibility coefficient, time elasticity coefficient, and state dependency coefficient; the task type includes: overall non-delayable task, overall delayable task, overall state-dependent task, divisible non-delayable task, divisible state-dependent task, or strong locality task. Based on the task type, the divisibility coefficient, temporal elasticity coefficient, state dependency coefficient, and spatial transferability coefficient are modified to obtain the modified divisibility coefficient, modified temporal elasticity coefficient, modified state dependency coefficient, and modified spatial transferability coefficient. The load spatiotemporal transfer potential coefficient is obtained by weighting and summing the descalability coefficient, temporal resilience coefficient, state dependence coefficient, and spatial transferability coefficient based on weighting factors, including: Based on the weighting factors, the modified descalability coefficient, modified time elasticity coefficient, modified state dependence coefficient, and modified spatial transferability coefficient are weighted and summed to obtain the load spatiotemporal transfer potential coefficient.

4. The method for assessing the spatiotemporal load transfer potential of a data center according to claim 3, characterized in that, Based on the task type, the descalability coefficient, temporal resilience coefficient, state dependency coefficient, and spatial transferability coefficient are modified to obtain the modified descalability coefficient, modified temporal resilience coefficient, modified state dependency coefficient, and modified spatial transferability coefficient, including: If the task type is a whole non-delayable task or a split non-delayable task, set the time elasticity coefficient to 0 to obtain the corrected time elasticity coefficient; otherwise, use the time elasticity coefficient as the corrected time elasticity coefficient. If the task type is an overall non-delayable task, an overall delayable task, or an overall state-dependent task, the divisibility coefficient is set to 0 to obtain the corrected divisibility coefficient; otherwise, the divisibility coefficient is used as the corrected divisibility coefficient. When the task type is a global state-dependent task, a separable state-dependent task, or a strongly local task, the state dependency coefficient is set to 1 to obtain the corrected state dependency coefficient, and the spatial transferability coefficient is set to 0 to obtain the corrected spatial transferability coefficient; otherwise, the state dependency coefficient is used as the corrected state dependency coefficient, and the spatial transferability coefficient is used as the corrected spatial transferability coefficient.

5. The method for assessing the spatiotemporal load transfer potential of a data center according to claim 3, characterized in that, Based on the divisibility coefficient, time flexibility coefficient, and state dependency coefficient, the task type is determined, including: If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be an overall non-delayable task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be an overall delayable task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, the time elasticity coefficient is less than the second preset time elasticity threshold, and the state dependence coefficient is not less than the preset state dependence threshold, the task type is determined to be an overall state-dependent task; wherein, the first preset time elasticity threshold is less than the second preset time elasticity threshold. If the divisibility coefficient is not less than the preset splitting threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is less than the preset state dependency threshold, the task type is determined to be a divisible but non-delayable task. If the divisibility coefficient is not less than the preset splitting threshold, the time elasticity coefficient is not less than the first preset time elasticity threshold, the time elasticity coefficient is less than the second preset time elasticity threshold, and the state dependency coefficient is not less than the preset state dependency threshold, the task type is determined to be a divisible state dependency task. If the divisibility coefficient is less than the preset divisibility threshold, the time elasticity coefficient is less than the first preset time elasticity threshold, and the state dependency coefficient is not less than the preset state dependency threshold, the task type is determined to be a strongly local task.

6. The method for assessing the spatiotemporal load transfer potential of a data center according to claim 1, characterized in that, Based on the latest completion deadline and the estimated execution time, the task slack is determined, including: Get the current evaluation time in real time; The remaining schedulable time for the task is determined based on the difference between the latest completion deadline and the current assessment time. The sum of the estimated execution time and the preset zero-positive number will be used as the baseline execution time for the task. The task slack is calculated as the ratio of the remaining schedulable time of the task to the baseline execution time of the task.

7. The method for assessing the spatiotemporal load transfer potential of a data center according to claim 1, characterized in that, The spatial transferability coefficient is determined based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance, including: Calculate the amount of data that the link can transmit within the scheduling period based on the available bandwidth of the link and the length of the scheduling period; The bandwidth adaptation factor is obtained by normalizing and truncating the ratio of the amount of data that the link can transmit to the amount of data required for task migration. The distance adaptation factor is calculated based on the data center transmission distance and the preset acceptable distance threshold. The spatial transferability coefficient is obtained by weighted summation of the bandwidth adaptation factor and the distance adaptation factor.

8. A device for assessing the potential for load transfer in a data center, characterized in that, include: The evaluation module includes a parameter acquisition module, a descalability calculation module, a time elasticity calculation module, a state dependency calculation module, a spatial transferability calculation module, a transfer potential coefficient calculation module, and a transfer potential evaluation module. The evaluation parameter acquisition module is used to acquire the load transfer evaluation parameters of the data center to be evaluated. The load transfer evaluation parameters include: maximum number of parallel subtasks, latest completion time limit, estimated execution time, total amount of migration status data, available link bandwidth, scheduling period length, amount of data required for task migration, and data center transmission distance. The divisibility calculation module is used to determine the divisibility coefficient based on the ratio of the maximum number of parallel subtasks to a preset parallelism benchmark value. The time elasticity calculation module is used to determine the task slack based on the latest completion deadline and the expected execution time, and to normalize and truncate the task slack to obtain the time elasticity coefficient. The state dependency calculation module is used to determine the state dependency coefficient based on the ratio of the total amount of migration state data to the preset state quantity benchmark value. The spatial transferability calculation module is used to determine the spatial transferability coefficient based on the available bandwidth of the link, the length of the scheduling period, the amount of data required for task migration, and the data center transmission distance. The transfer potential coefficient calculation module is used to perform a weighted summation of the decomposability coefficient, time elasticity coefficient, state dependence coefficient, and spatial transferability coefficient to obtain the load spatiotemporal transfer potential coefficient. The load transfer potential assessment module is used to determine the load transfer potential of the data center to be assessed based on the load transfer potential coefficient.

9. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the data center load spatiotemporal shift potential assessment method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the data center load spatiotemporal transfer potential assessment method as described in any one of claims 1-7.

Images