Computing power network task scheduling method, task processing method, device and system

Abstract

The application discloses a computing power network task scheduling method, a task processing method, a device and a system. The method comprises the following steps: receiving a computing power service request of a user; decomposing an original request task into a plurality of atomic tasks and constructing an atomic task sequence; dividing the atomic task sequence into a plurality of time-sharing task sequences for sequential execution; inputting resource requirements and task attribute information of each atomic task under the current time-sharing task sequence, computing power states of each computing power node in the current global resource pool, an access power carbon emission factor and an access power price into a pre-trained matching model to obtain a plurality of computing power nodes with the lowest power consumption cost matched with each atomic task under the current time-sharing task sequence; and scheduling each atomic task under the current time-sharing task sequence to a corresponding computing power node. The embodiment of the application can perform task scheduling with green computing power as a priority on the premise of meeting the computing power service request of the user, and realize the lowest computing power cost.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of computing power network, and particularly relates to a computing power network task scheduling method, a task processing method, a device and a system. BACKGROUND

[0002] At present, computing power has been widely integrated into various aspects of social production and life, providing basic driving force for digital transformation of thousands of industries, and is the core digital resource in today's society. However, data centers consume a huge amount of electricity, and with the rapid growth of computing power, power consumption and carbon emissions also increase, so the country actively promotes the construction of green data centers. On the one hand, it speeds up the energy-saving reconstruction of data centers and improves the power usage effectiveness (PUE) of data centers. On the other hand, it improves the level of green energy power such as water power, wind power and solar power used by data centers, and reduces carbon emissions.

[0003] Although green energy power has the advantages of low cost and low carbon emissions, it also has unpredictability and high volatility, and other grid power needs to be introduced to ensure stability. Therefore, how to prioritize green computing power for task scheduling to achieve the goal of minimizing computing power cost under the premise of meeting user computing power service requests has become a research topic. SUMMARY

[0004] The embodiments of the application provide a computing power network task scheduling method, a task processing method, a device and a system, which can prioritize green computing power for task scheduling to achieve the goal of minimizing computing power cost under the premise of meeting user computing power service requests.

[0005] In a first aspect, the embodiments of the application provide a computing power network task scheduling method for scheduling a server, which comprises:

[0006] receiving a computing power service request of a user;

[0007] decomposing an original request task into a plurality of atomic tasks and constructing an atomic task sequence;

[0008] dividing the atomic task sequence into a plurality of time-sharing task sequences to be executed in sequence, and the execution period of each time-sharing task sequence coincides with an electricity price adjustment period;

[0009] inputting, in a time sequence, resource demand and task attribute information of each atomic task under a current time-sharing task sequence, computing power state of each computing power node in a current global resource pool, access grid power carbon emission factor and access power price into a pre-trained matching model to obtain a plurality of computing power nodes with the lowest power consumption cost matched with each atomic task under the current time-sharing task sequence;

[0010] Scheduling each atomic task in the current time-sharing task sequence to a corresponding computing power node.

[0011] In some possible implementation manners of the first aspect, the splitting the atomic task sequence into a plurality of time-sharing task sequences executed sequentially comprises: estimating a total execution duration of the atomic task sequence; calculating a first ratio of the total execution duration to a duration of the electricity price adjustment period; rounding up the first ratio to obtain m; and splitting the atomic task sequence into m time-sharing task sequences executed sequentially; wherein a starting execution time of an i-th time-sharing task sequence is T0+(i-1)XT, an ending execution time of the i-th time-sharing task sequence is T0+iXT, m≥i≥1, T0 is a starting time of the electricity price adjustment period, and T is the duration of the electricity price adjustment period.

[0012] In some possible implementation manners of the first aspect, the splitting the atomic task sequence into a plurality of time-sharing task sequences executed sequentially comprises: for each atomic task in the atomic task sequence, determining whether there is a serial atomic task to be executed in front of the atomic task; in a case where there is no serial atomic task to be executed in front of the atomic task, calculating a second ratio of an estimated execution duration of the atomic task to the duration of the electricity price adjustment period; rounding up the second ratio to obtain n; in a case where n is equal to 1, not splitting the atomic task, and the atomic task is located in the first time-sharing task sequence; and in a case where n is greater than 1, splitting the atomic task, and the split atomic task is located in the first time-sharing task sequence to the n-th time-sharing task sequence.

[0013] In some possible implementation manners of the first aspect, before receiving the computing power service request of the user, the method further comprises: receiving computing network resource registration information sent by a resource pool side server; and marking, according to the computing network resource registration information, a computing power state, an access power grid power carbon emission factor, and an access power price of each computing power node in the resource pool; wherein the computing power state comprises a plurality of parameters selected from the following: location, service level, power usage efficiency, computing power, computing efficiency, computing power architecture, and network bandwidth.

[0014] In some possible implementation manners of the first aspect, the method further comprises: in a case where the computing network resource registration information has been updated, updating the computing power state, the access power grid power carbon emission factor, and the access power price of each computing power node in the resource pool that are marked.

[0015] In a second aspect, an embodiment of the present application provides a task processing method, used for a resource pool side server, and the method comprises the following steps.

[0016] processing computing power state information, an access power grid power carbon emission factor, and an access power price of a local resource pool to generate computing network resource registration information;

[0017] sending the computing network resource registration information to a scheduling server;

[0018] Receive the atomic task and the computing power node information sent by the scheduling server;

[0019] According to the atomic task and the computing power node information, a computing network resource template is arranged for the atomic task;

[0020] The computing network resource template is sent to the designated computing power node, and the computing power node is instantiated according to the network resource template. After the atomic task processing is completed or the running time arrives, the computing result is returned and the computing power node resource is released.

[0021] In some possible implementation manners of the second aspect, the method further includes: monitoring the computing power state information of the local resource pool, the access power grid power carbon emission factor and the access power price; updating the computing network resource registration information of the local resource pool when any one of the computing power state information of the local resource pool, the access power grid power carbon emission factor and the access power price changes; and synchronizing the updated computing network resource registration information to the scheduling server.

[0022] In a third aspect, an embodiment of the present application provides a computing power network task scheduling device, configured for a scheduling server, and the device includes:

[0023] The receiving module is configured to receive a computing power service request of a user;

[0024] The deconstruction module is configured to decompose the original request task into a plurality of atomic tasks and construct an atomic task sequence;

[0025] The segmentation module is configured to segment the atomic task sequence into a plurality of time-sharing task sequences that are executed in sequence, and the execution period of each time-sharing task sequence coincides with a power price adjustment period;

[0026] The matching module is configured to input, in a time sequence, the resource requirement and the task attribute information of each atomic task under a current time-sharing task sequence, the computing power state of each computing power node in a current global resource pool, the access power grid power carbon emission factor and the access power price into a pre-trained matching model to obtain a plurality of computing power nodes with the lowest electricity cost that match each atomic task under the current time-sharing task sequence;

[0027] The scheduling module is configured to schedule each atomic task under the current time-sharing task sequence to a corresponding computing power node.

[0028] In a fourth aspect, an embodiment of the present application provides a task processing device, configured for a resource pool side server, and the device includes:

[0029] The generating module is configured to process the computing power state information of the local resource pool, the access power grid power carbon emission factor and the access power price to generate computing network resource registration information;

[0030] The first sending module is configured to send the computing resource registration information to the scheduling server.

[0031] The receiving module is configured to receive the atomic task and the computing resource node information sent by the scheduling server.

[0032] The arrangement module is configured to arrange a computing resource template for the atomic task according to the atomic task and the computing resource node information.

[0033] The second sending module is configured to send the computing resource template to the designated computing resource node.

[0034] The computing resource node is configured to be instantiated according to the computing resource template, and return the calculation result and release the computing resource node resource after the atomic task processing is completed or the running time arrives.

[0035] In a fifth aspect, an embodiment of the present application provides a computing resource network task scheduling system, which comprises a scheduling server, a cloud, an edge, an end resource pool, and a plurality of resource pool side servers deployed in the cloud, the edge, and the end resource pool; the scheduling server comprises the computing resource network task scheduling apparatus; and the resource pool side server comprises the task processing apparatus.

[0036] In a sixth aspect, an embodiment of the present application provides a terminal device, which comprises a processor and a memory storing computer program instructions; the processor executes the computer program instructions to implement the computing resource network task scheduling method according to any one of claims 1-5, or implement the task processing method according to claim 6 or 7.

[0037] In a seventh aspect, an embodiment of the present application provides a computer readable storage medium, which stores computer program instructions; the computer program instructions are executed by a processor to implement the computing resource network task scheduling method or the task processing method.

[0038] As described above, the computing resource network task scheduling method in the embodiment of the present application, after receiving the computing resource service request of the user, first disassembles the original request task into a plurality of atomic tasks and constructs an atomic task sequence; then divides the atomic task sequence into a plurality of time-sharing task sequences to be executed in sequence, and the execution period of each time-sharing task sequence coincides with one electricity price adjustment period; then inputs the resource demand and the task attribute information of each atomic task in the current time-sharing task sequence, the computing resource state of each computing resource node in the current global resource pool, the access power grid power carbon emission factor, and the access power price into a pre-trained matching model in time sequence, to obtain a plurality of computing resource nodes with the lowest electricity cost matched with each atomic task in the current time-sharing task sequence; and finally schedules each atomic task in the current time-sharing task sequence to the corresponding computing resource node.

[0039] On the one hand, since the embodiment of the application is to schedule tasks in units of time-sharing task sequences, and the electricity price remains unchanged within the execution period of a time-sharing task sequence, the influence of the change of electricity price with time and region on the task scheduling of the computing power network can be excluded, and the accuracy of the task scheduling of the computing power network is improved. On the other hand, since the embodiment of the application includes the access grid electricity carbon emission factor into the computing power node matching model algorithm, the matching model takes the lowest power cost as the training target, and under the condition of the same power cost, the smaller the access grid electricity carbon emission factor, the higher the priority of the selected computing power node, so that the computing power task scheduling can be adaptively performed under the premise of meeting the user computing power service request, and the purpose of preferentially using green computing power and achieving the lowest power cost is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed to be used in the embodiments of the application will be briefly introduced as follows. Those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0041] Figure 1 One of the structural schematic diagrams of the computing power network task scheduling system provided by the embodiments of the application;

[0042] Figure 2 One of the flow schematic diagrams of the computing power network task scheduling method provided by the embodiments of the application;

[0043] Figure 3 One of the segmentation schematic diagrams of the atomic task provided by the embodiments of the application;

[0044] Figure 4 The second segmentation schematic diagram of the atomic task provided by the embodiments of the application;

[0045] Figure 5 The flow schematic diagram of the task processing method provided by the embodiments of the application;

[0046] Figure 6 The second structural schematic diagram of the computing power network task scheduling system provided by the embodiments of the application;

[0047] Figure 7 The second flow schematic diagram of the computing power network task scheduling method provided by the embodiments of the application;

[0048] Figure 8 The structural schematic diagram of the computing power network task scheduling device provided by the embodiments of the application;

[0049] Figure 9 The structural schematic diagram of the task processing device provided by the embodiments of the application;

[0050] Figure 10 is a structural schematic diagram of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0051] The features and exemplary embodiments of various aspects of the present application will be described below in detail, in order to make the purposes, technical solutions and advantages of the present application more clear and apparent, the present application will be further described in detail below in combination with the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, but not to limit the present application. The present application can be implemented without some of these specific details by those skilled in the art. The following description of the embodiments is only to provide a better understanding of the present application by showing examples of the present application.

[0052] It should be noted that, in this paper, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the elements defined by the statement "include" do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0053] Computing power network is a new type of information infrastructure that allocates and flexibly schedules computing resources, storage resources and network resources according to business needs among cloud, edge and end. The essence of computing power network is a computing power resource service. In the future, enterprise customers or individual users not only need network and cloud, but also need to flexibly schedule computing tasks to the right place.

[0054] The computing power network data center uses green energy power such as water, wind and solar power to reduce carbon emissions. However, green energy power has unpredictability and high volatility. Not only does the electricity price change dynamically, but also the electricity price in different regions is different. Therefore, it is necessary to introduce other grid power to ensure stability. Therefore, it is difficult to meet the requirement of maximizing the use of green computing power to reduce computing power cost by using fixed rule scheduling strategy.

[0055] To solve this problem, the inventors of the present application propose that a green computing power priority adaptive scheduling strategy can be used to achieve the goal of minimizing computing power cost.

[0056] As Figure 1As shown, this application provides a computing power network task scheduling system. The system includes a scheduling server F0, cloud, edge, and terminal resource pools, and multiple resource pool-side servers (F11, F12, F13, FN1, FN2, and FN3) deployed in the cloud, edge, and terminal resource pools. The scheduling server F0 executes the computing power network task scheduling method of this application embodiment, also referred to as the computing power network service subsystem. Each resource pool-side server executes the task processing method of the application embodiment, also referred to as the computing power network resource orchestration subsystem. The computing power network service subsystem can maintain communication and interaction with the computing power network resource orchestration subsystems located on each resource pool side.

[0057] Several central cloud resource pools, edge cloud resource pools, and endpoint resource pools can form a global computing power resource pool. Generally, central cloud resource pools have ultra-large-scale computing power, lower power usage efficiency (PUE) through centralized management, and can access green electricity such as wind and hydropower on a larger scale. End-side resource pools, being closer to customers, have advantages in low latency and low power consumption due to network communication, but their computing power is limited and their PUE is relatively high, with less available green electricity and greater volatility. Edge cloud resource pools have computing power and network latency between the two, with a PUE that is relatively higher than that of central cloud but lower than that of endpoint resource pools, and their power consumption is similar to that of central cloud resource pools.

[0058] The following is a detailed description of the computing power network task scheduling method provided in the embodiments of this application.

[0059] like Figure 2 As shown, the computing power network task scheduling method provided in this application embodiment includes steps S201 to S205.

[0060] S201, Receive user's computing power service request.

[0061] S202. Decompose the original request task into multiple atomic tasks and construct an atomic task sequence.

[0062] S203. Divide the atomic task sequence into multiple time-sharing task sequences to be executed sequentially.

[0063] The execution cycles of each time-of-use task sequence overlap within a single electricity price adjustment cycle, which is the period between the end of the previous electricity price adjustment and the start of the next. For example, if electricity prices are adjusted at 0:00, 4:00, 8:00, 12:00, and so on, then the electricity price adjustment cycles are 0:00–4:00, 4:00–8:00, 8:00–12:00, and so on. Accordingly, the execution cycles of multiple time-of-use task sequences can be set sequentially to 0:00–4:00, 4:00–8:00, 8:00–12:00, etc. The electricity price remains unchanged within the execution cycle of a time-of-use task sequence, and different execution cycles of different time-of-use task sequences can correspond to different electricity prices.

[0064] S204, inputting the resource requirement and task attribute information of each atomic task under the current time-sharing task sequence, the computing power state of each computing power node in the current global resource pool, the access grid power carbon emission factor and the access power price into the pre-trained matching model in chronological order, to obtain a plurality of computing power nodes with the lowest electricity cost matched with each atomic task under the current time-sharing task sequence.

[0065] S205, scheduling each atomic task under the current time-sharing task sequence to the corresponding computing power node.

[0066] The resource information required by each atomic task can include computing resources, memory resources, network resources and storage resources required by the atomic task, and the task attribute information can include network delay, SLA service level, task duration, atomic task affinity group and atomic task anti-affinity group required by the atomic task, and the resource requirement information and the task attribute information are used together to represent the user computing power task.

[0067] The computing power state of each computing power node can include multiple parameters: location, service level, power usage efficiency, computing power, computing efficiency, computing power architecture and network bandwidth. The computing efficiency refers to the computing power efficiency of IT computing equipment, indicating the computing power scale that can be generated per unit energy consumption, generally in Flops / W; the power usage efficiency refers to the ratio of total energy consumption of the data center to IT load energy consumption, wherein the total energy consumption of the data center includes IT device energy consumption and energy consumption of cooling, power distribution and other systems, and the value is greater than 1, and the closer to 1, the less non-IT device energy consumption, i.e. the better energy efficiency level.

[0068] The access grid power carbon emission factor of each computing power node refers to the carbon emission corresponding to unit energy consumption, which is used to represent the green power usage of each computing power node, and the carbon emission factor is generally in tCO2 / MWh, and the lower the access grid power carbon emission factor, the more green power usage.

[0069] The access power price of each computing power node is used to represent the electricity fee of the access computing power node, and the lower the access power price, the lower the electricity cost.

[0070] In some embodiments, a matching model can be pre-constructed, and then a plurality of computing power tasks are obtained. According to the method in steps S202 and S203, each computing power task is first decomposed into an atomic task sequence, and then the atomic task sequence is segmented into a plurality of time-sharing task sequences. For each time-sharing task sequence, each atomic task in the time-sharing task sequence is pre-assigned a computing power node. Then, based on the lowest power consumption cost of the entire time-sharing task sequence execution cycle as the training target, the resource demand and task attribute information of each atomic task in the time-sharing task sequence, the computing power state of each computing power node in the current global resource pool, the access power grid power carbon emission factor, the access power price, and the pre-assigned computing power node are input into the pre-constructed matching model for deep reinforcement learning training to obtain the pre-trained matching model.

[0071] As described above, in the computing power network task scheduling method in the embodiments of the present application, after receiving the user's computing power service request, the original request task is first disassembled into a plurality of atomic tasks and an atomic task sequence is constructed. Then, the atomic task sequence is segmented into a plurality of time-sharing task sequences for sequential execution, and the execution cycle of each time-sharing task sequence coincides with a power price adjustment cycle. Then, in time sequence, the resource demand and task attribute information of each atomic task in the current time-sharing task sequence, the computing power state of each computing power node in the current global resource pool, the access power grid power carbon emission factor, and the access power price are input into the pre-trained matching model to obtain a plurality of computing power nodes matched with each atomic task in the current time-sharing task sequence with the lowest power consumption cost. Finally, each atomic task in the current time-sharing task sequence is scheduled to the corresponding computing power node.

[0072] On the one hand, since the embodiments of the present application perform task scheduling in units of time-sharing task sequences, and the power price remains unchanged within the execution cycle of a time-sharing task sequence, the influence of the change of power price over time and region on the computing power network task scheduling can be excluded, and the accuracy of the computing power network task scheduling can be improved. On the other hand, since the embodiments of the present application include the access power grid power carbon emission factor in the computing power node matching model algorithm, and the matching model takes the lowest power cost as the training target, under the condition of the same power cost, the smaller the access power grid power carbon emission factor, the higher the priority of the computing power node selected, so that the computing power task scheduling can be adaptively performed under the premise of meeting the user's computing power service request, the green computing power is preferentially used, and the purpose of achieving the lowest power cost is achieved.

[0073] The specific implementation of each step is described below.

[0074] In some embodiments, in S202, the original request task can be disassembled into multiple atomic tasks and an atomic task sequence is constructed according to the task type contained in the original request task or the execution relationship between tasks, and the application does not limit the disassembly manner of the task. By structuring the user's original demand into an atomic task group, the normalized computing power demand of different application services is realized, and the task scheduling can be more refined.

[0075] In some embodiments, in S203, the atomic task sequence can be divided into multiple time-sharing task sequences for sequential execution by the following method.

[0076] First, considering the execution period of each time-sharing task sequence coincides with a price adjustment period, the total execution time of the atomic task sequence can be estimated first with respect to the execution period of the time-sharing task sequence; then a first ratio of the total execution time and the price adjustment period time is calculated; then m is obtained by rounding up the first ratio; then the atomic task sequence is divided into m time-sharing task sequences for sequential execution; wherein the starting execution time of the i-th time-sharing task sequence is T0+(i-1)×T, and the ending execution time is T0+i×T, m≥i≥1, T0 is the starting time of the price adjustment period, and T is the price adjustment period time.

[0077] Since some atomic tasks in the atomic task sequence are executed in parallel, and some atomic tasks need to wait for the execution of other atomic tasks to be executed, which is also called serial atomic tasks. Therefore, for each atomic task in the atomic task sequence, it is necessary to determine whether there is a serial atomic task to be executed in front of the atomic task before dividing.

[0078] In some embodiments, if there is no serial atomic task to be executed in front of the atomic task, it is determined that the atomic task can be located in the first time-sharing task sequence; next, a second ratio of the estimated execution time of the atomic task and the price adjustment period time can be calculated; and n is obtained by rounding up the second ratio.

[0079] In the case where n is equal to 1, it is indicated that the execution period of the atomic task is short, shorter than the price adjustment period, so that the atomic task does not need to be divided, and the atomic task is located in the first time-sharing task sequence.

[0080] In the case where n is greater than 1, it is indicated that the execution period of the atomic task is long, which needs to span multiple price adjustment periods, in order to avoid the influence of price changes on task scheduling, the atomic task needs to be divided, and the divided atomic task can be located in the first to the nth time-sharing task sequence.

[0081] In some embodiments, if the atomic task is preceded by a serial atomic task to be executed, the atomic task needs to wait for the execution of the preceding serial atomic task to be completed before starting execution. In this case, if the execution time of the last time-sharing task sequence of the serial atomic task is equal to the price adjustment period, the atomic task can be located in the next time-sharing task sequence after the execution of the serial atomic task is completed; if the execution time of the last time-sharing task sequence of the serial atomic task is shorter than the price adjustment period, the atomic task can be located in the last time-sharing task sequence of the serial atomic task together with the serial atomic task to fill the last time-sharing task sequence of the serial atomic task; or the atomic task can be directly located in the next time-sharing task sequence after the execution of the serial atomic task is completed, and execution starts from the next time-sharing task sequence. Those skilled in the art can select a suitable segmentation method according to the actual situation, which is not limited here.

[0082] The following will be described in combination with Figure 3 and Figure 4 The segmentation process of the atomic task sequence is illustrated by way of example.

[0083] Figure 3 Atomic tasks R1 and R2 are shown in FIG. 1, R1 is used for video encoding, and R2 is used for uploading the video to a video website. R2 needs to wait for the execution of R1 to be completed before starting execution, that is, R1 and R2 are serial atomic tasks. It is assumed that the price adjustment period is 4h,

[0084] In which, R1 is expected to consume 14h, R1 can be segmented into 4 time-sharing task sequences, the execution time of R1 in the first 3 time-sharing task sequences is 4h, and the execution time of R1 in the 4th time-sharing task sequence is 2h. R2 is expected to consume 2h, R2 can not be segmented, but R2 is divided into the 4th time-sharing task sequence.

[0085] Figure 4 Atomic tasks R3, R4 and R5 are shown in FIG. 2, R3 is used for encrypting the video, R4 is used for crushing the video after R1 is completed, and R5 is used for executing a crawler task, that is, R4 and R3 are serial atomic tasks, and R5 and R3 are parallel atomic tasks.

[0086] In which, R3 is expected to consume 6h, R3 can be segmented into 2 time-sharing task sequences, the execution time of R3 in the first time-sharing task sequence is 4h, and the execution time of R3 in the 2nd time-sharing task sequence is 2h. R4 is expected to consume 1.5h, R4 can not be segmented, but R4 is divided into the 2nd time-sharing task sequence. R5 is expected to consume 24h, R1 can be segmented into 6 time-sharing task sequences.

[0087] In some embodiments, before S201, the computing power network task scheduling method in the embodiments of the present application can further include: receiving the computing power network resource registration information sent by the resource pool side server; and marking the computing power state, the access power grid power carbon emission factor and the access power price of each computing power node in the resource pool according to the computing power network resource registration information.

[0088] In an example, the format of the computing power network resource registration information is as follows:

[0089] {resourceID: "resourceID", resourceDomain: "resourceDomain", resourcePUE: "resourcePUE", resourceSLA: "resourceSLA", resourceLocation: "resourceLocation", resourceStatus: "{{totalCPU: "totalCPU", totalMEM: "totalMEM", totalBW: "totalBW"}, {availableCPU: "availableCPU", availableMEM: "availableMEM", availableBW: "availableBW"}}", nodeVec: "{{nodeIP: "nodeIP", nodeCE: "nodeCE", powerCEF: "powerCEF", powerPrice: "powerPrice", Status: "{{totalCPU: "totalCPU", totalMEM: "totalMEM", totalBW: "totalBW"}, {availableCPU: "availableCPU", availableMEM: "availableMEM", availableBW: "availableBW"}}"...}}

[0090] Wherein, resourceID is the resource pool ID, resourceDomain is the resource pool domain, resourcePUE is the resource pool power usage efficiency, resourceSLA is the resource pool service level, resourceLocation is the resource pool location, resourceStatus is the resource pool resource status, the total and allocable amount of CPU, MEM and network bandwidth resources, N groups of computing power nodes Node (at least containing 1 effective computing power node), nodeIP is the computing power node IP, nodeCE is the computing power node computing efficiency, powerCEF is the current access power grid power carbon emission factor of the computing power node, powerPrice is the current power price of the computing power node, Status is the computing power node resource status, and the total and allocable amount of CPU, MEM and network bandwidth resources.

[0091] In some embodiments, in the case that the computing network resource registration information has been updated, the computing power state, the access power grid power carbon emission factor and the access power price of the computing power node in the currently marked resource pool also need to be updated, so as to ensure the accuracy of the matching model input data and avoid reducing the scheduling accuracy of the computing power network task due to inaccurate data.

[0092] As shown in Figure 5 The embodiments of the present application also provide a task processing method, which is used for a resource pool side server and includes steps S501 to S505.

[0093] S501, processing the computing power state information, the access power grid power carbon emission factor and the access power price of the local resource pool to generate computing network resource registration information.

[0094] S502, sending the computing network resource registration information to a scheduling server.

[0095] S503, receiving atomic tasks and computing power node information sent by the scheduling server.

[0096] S504, scheduling a computing network resource template for the atomic tasks according to the atomic tasks and the computing power node information.

[0097] S505, sending the computing network resource template to a designated computing power node, and instantiating the computing network resource template by the computing power node; after the atomic task processing is completed or the running time arrives, returning the calculation result and releasing the computing power node resource.

[0098] In some embodiments, to avoid the arrival of the resource pool aging time or update, the task processing method in the embodiments of the application further includes: monitoring the computing power state information of the local resource pool, the access grid power carbon emission factor and the access power price; updating the grid resource registration information of the local resource pool when any one of the computing power state information of the local resource pool, the access grid power carbon emission factor and the access power price changes; and synchronizing the updated grid resource registration information to the scheduling server.

[0099] For the convenience of those skilled in the art, the technical solutions of the embodiments of the application are illustrated below in conjunction with Figure 6 and Figure 7 the embodiments of the application.

[0100] Figure 6 The illustrated computing power network task scheduling system includes a computing power network service subsystem and a computing network resource arrangement subsystem and related network devices, wherein the computing power network service subsystem can include a computing network resource management module, a computing network demand disassembly module and a computing network task scheduling module; the computing network resource arrangement subsystem can include a computing network resource perception module, a computing network resource arrangement module, a computing network task management module and a power resource management module.

[0101] Figure 7 The illustrated computing power network task scheduling flow is based on the system architecture. Figure 6

[0102] S701, system initialization.

[0103] First, the computing power network service subsystem initialization and the computing network resource arrangement subsystem initialization, wherein the computing power network service subsystem initialization includes loading initialization configuration information and accessing resource pool preset static information, initializing each sub-module to wait for receiving a request, the computing network resource arrangement subsystem initialization includes the computing network resource perception module acquiring the computing power and network resource information of the local resource pool and normalizing the measurement, the resource pool PUE, the computing power node CE, the access grid power carbon emission factor mark, the power price timetable registered to the computing power network service subsystem, and real-time monitoring of the local resource pool state periodically synchronized to the computing power network service subsystem.

[0104] Second, the resource management module of the computing power network service subsystem receives the grid resource registration information of the computing power resource pool, identifies the computing power according to the resource attribute, identifies the location, the SLA level, the computing power, the network bandwidth, the computing power architecture (CPU, GPU, etc.), the computing power state characteristics of the resource pool PUE, the computing power node CE, the access grid power carbon emission factor and the access power price;

[0105] S702, the computing power network service subsystem receives the client request and structures the original task of the client into N atomic tasks and constructs an atomic task group Vec(S1,..., S​N ), further according to the power price adjustment period Δt, the atomic task group is divided into M (M≥1) time sequences Vec t1 ...Vec tM , also known as a time-sharing task sequence.

[0106] wherein the resource information required by each atomic task is (C k ,M k ,N k ,S k ), respectively, the computing, memory, network, and storage resources required by the Kth atomic task; the atomic task attribute information is (L k ,SLA k ,T k ,A k (S i ,...), AA k (S j ,...)), respectively, the network delay, SLA service level, task duration, atomic task affinity group, and atomic task anti-affinity group of the Kth atomic task; and then the atomic task group is added to the task processing queue. For the atomic task S k There can be only one time interval or several time intervals, and in some cases, if all N atomic tasks end within one time interval Δt, it is not necessary to perform time sequence segmentation on them;

[0107] S703, the computing network task scheduling module of the computing network service subsystem sequentially matches, according to the time sequence, the atomic task group resource demand and task attribute requirement of the current time sequence, the current global resource pool computing power state, the current resource pool access power grid power carbon emission factor, and the access power price using a deep reinforcement learning model, to match the N best computing nodes Vec ti (S1,..., S N ) of the time sequence Vec ti (Node1,..., Node N ).

[0108] In the embodiments of the present application, the consumed electric energy of the time sequence Vec ti (S1,..., S N ) can be expressed as:

[0109] E{Vec ti}=∑{E Sj +ΔE sj}(j∈(1, N)) (1)

[0110] wherein E Sj represents the electric energy required for completing the jth atomic task of the current time sequence, and ΔE SjΔE represents the electric energy required to migrate the jth atomic task of the current time sequence to other computing nodes sj The size depends on the atomic task S j The migration distance, the migration data volume, and the complexity of the service itself can be intelligently predicted using machine learning. If no computing node migration occurs or j = 1, ΔE sj = 0.

[0111] For the atomic task S j , the computing power required to complete the task is fixed, but the size of the electric energy consumed is related to the resource pool PUE and the computing efficiency of the computing node. The lower the resource pool PUE and the greater the computing efficiency of the computing node, the less electric energy is consumed. The calculation formula of the electric energy consumed is as follows:

[0112]

[0113] Wherein, W PUE is the resource pool electric energy use efficiency normalization weight, whose value is greater than or equal to 1, W CE is the computing node computing efficiency normalization weight, F Sj is the standard electric energy required to complete the atomic task S j .

[0114] The current time sequence atomic task group power cost can be represented as:

[0115] C{Vec ti}=∑{C sj}(j∈(1,N)) (3)

[0116] Wherein, C sj is the electric power cost of the current time sequence atomic task S j , C sj == W CEF * P sj * E Sj , W CEF is the resource pool access grid power carbon emission factor normalization weight coefficient, the smaller the value, the less carbon emission, P sj is the current real-time power price. Under the same power cost, the smaller the carbon emission factor, the higher the priority of the computing node selected.

[0117] Due to the unpredictability and high volatility of green power such as wind power, solar power, and hydropower, the dynamic change of power price, and the different power prices in different regions, the traditional fixed rule-based scheduling strategy cannot meet the maximum use of green computing power to reduce computing power cost. Here, a deep reinforcement learning scheduling strategy is used to adaptively schedule computing power tasks based on global computing power state, user computing power tasks, and resource pool access power information.

[0118] S704, the computing network service subsystem's computing network task scheduling module will schedule the current time-series atomic task group Vec ti (S1, ...,S) N They are respectively scheduled to the corresponding resource pool computing power nodes Vec ti (Node1,...,Node N It performs resource orchestration and instantiation, and returns the calculation results and releases node resources after the atomic task is completed or the runtime is reached.

[0119] Among them, the computing network resource orchestration module of the computing network service subsystem, based on the current time series atomic task group Vec ti (S1, ...,S) N Orchestrate atomic tasks S j The computing network resource template, the computing network task management module of the computing power network service subsystem, according to the atomic task S on the computing power nodes of the specified resource pool. j The computing network resource template is instantiated, and the computing node resources are released after the atomic task is completed or the running time is reached.

[0120] S705. Determine whether all M time series have been processed. If yes, end the process; otherwise, return to S703.

[0121] As mentioned above, on the one hand, by introducing the power usage efficiency of the data center resource pool, the energy efficiency of computing nodes, and the carbon emission factor identifier of access electricity, this application enables computing network tasks to be prioritized for green computing power scheduling in the global resource pool, thereby reducing computing power costs and carbon emissions. On the other hand, by dividing computing tasks into multiple time series according to the electricity price adjustment cycle and scheduling them separately based on a deep reinforcement learning model, this approach can maximize the adaptation to the high volatility of green energy electricity, dynamic changes in electricity prices, and different electricity prices in different regions, thereby achieving the lowest electricity cost to meet the computing task processing requirements.

[0122] like Figure 8 As shown in the figure, this application embodiment also provides a computing power network task scheduling device 800 for scheduling servers. The computing power network task scheduling device 800 includes: a receiving module 801, a deconstruction module 802, a segmentation module 803, a matching module 804, and a scheduling module 805.

[0123] The receiving module 801 is used to receive users' computing power service requests;

[0124] Destructuring module 802 is used to break down the original request task into multiple atomic tasks and construct a sequence of atomic tasks;

[0125] The segmentation module 803 is configured to segment the atomic task sequence into a plurality of time-sharing task sequences executed in sequence, and a period of execution of each time-sharing task sequence coincides with one electricity price adjustment period.

[0126] The matching module 804 is configured to input, in a time sequence, resource demand and task attribute information of each atomic task in a current time-sharing task sequence, computing power state of each computing power node in a current global resource pool, access power grid power carbon emission factor, and access power price into a pre-trained matching model to obtain a plurality of computing power nodes with the lowest power consumption cost matched with each atomic task in the current time-sharing task sequence.

[0127] The scheduling module 805 is configured to schedule each atomic task in the current time-sharing task sequence to a corresponding computing power node.

[0128] It should be noted that the computing power network task scheduling apparatus 800 is an apparatus corresponding to the computing power network task scheduling method applied to the scheduling server, and all implementation manners in the above method embodiments are applicable to the embodiments of the apparatus, and the same technical effects can be achieved.

[0129] As shown in Figure 9 The present application also provides a task processing apparatus 900 for a resource pool side server, which comprises a generation module 901, a first sending module 902, a receiving module 903, an arrangement module 904, a second sending module 905, and a computing power node 906.

[0130] The generation module 901 is configured to process computing power state information of a local resource pool, access power grid power carbon emission factor, and access power price to generate computing network resource registration information.

[0131] The first sending module 902 is configured to send the computing network resource registration information to a scheduling server.

[0132] The receiving module 903 is configured to receive atomic tasks and computing power node information sent by the scheduling server.

[0133] The arrangement module 904 is configured to arrange a computing network resource template for the atomic tasks according to the atomic tasks and the computing power node information.

[0134] The second sending module 905 is configured to send the computing network resource template to a designated computing power node.

[0135] The computing power node 906 is configured to be instantiated according to the network resource template, and return a calculation result and release computing power node resources after the atomic task processing is completed or the running time arrives.

[0136] It should be noted that the task processing apparatus 900 is an apparatus corresponding to the task processing method applied to the resource pool side server described above, and all implementation manners in the method embodiments are applicable to the embodiments of the apparatus, and the same technical effects can also be achieved.

[0137] Figure 10 A hardware structure schematic diagram of a terminal device provided in an embodiment of the present application is shown.

[0138] The terminal device can include a processor 1001 and a memory 302 storing computer program instructions.

[0139] Specifically, the processor 1001 can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or can be configured as one or more integrated circuits that implement one or more embodiments of the present application.

[0140] The memory 1002 can include a mass storage for data or instructions. By way of example and not limitation, the memory 1002 can include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 1002 can include removable or non-removable (or fixed) media. Where appropriate, the memory 1002 can be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, the memory 1002 is non-volatile solid-state memory.

[0141] The processor 1001 reads and executes the computer program instructions stored in the memory 1002 to implement any one of the computing power network task scheduling method and the task processing method in the above embodiments.

[0142] In one example, the terminal device can also include a communication interface 1003 and a bus 1010. As shown, the processor 1001, the memory 1002, and the communication interface 1003 are connected through the bus 1010 and complete communication among each other. Figure 10

[0143] The communication interface 1003 is mainly used to realize the communication between each module, apparatus, unit and / or device in the embodiments of the present application.

[0144] ​Bus 1010 includes hardware, software, or both, to couple components of the online data traffic metering device to each other and to other components. While the application is not limited to particular bus structures, in this application, bus could be, for example, an accelerated graphics port (AGP) or other graphics bus, an enhanced industry standard architecture (EISA) bus, a front-side bus (FSB), a HyperTransport (HT) interconnect, an industry standard architecture (ISA) bus, an infiniband (IB) interconnect, a low pin count (LPC) bus, a memory bus, a microchannel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standards board (VLB) bus, or another suitable bus or a combination of two or more of these. Bus 1010 can include one or more buses, where appropriate. Although this application describes and illustrates a particular bus, this application contemplates any suitable bus or interconnect.

[0145] In addition, in combination with the computing power network task scheduling method and the task processing method in the above embodiments, the embodiments of the present application can provide a computer storage medium for implementation. The computer storage medium has computer program instructions stored thereon; the computer program instructions are executed by a processor to implement any one of the computing power network task scheduling method and the task processing method in the above embodiments.

[0146] It is to be understood that the application is not limited to the particular configurations and processes described and illustrated herein. Detailed descriptions of known methods are omitted so as not to obscure the application in unnecessary detail. In the above embodiments, several specific steps are described and illustrated as examples. However, the method processes of the application are not limited to the specific steps described and illustrated, and one skilled in the art can make various changes, modifications and additions, or change the order of the steps, after understanding the spirit of the application.

[0147] The functional blocks shown in the above structural block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the application are program or code segments used to perform the required tasks. The program or code segments can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. The "machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. The code segments can be downloaded via a computer network such as the Internet, an intranet, etc.

[0148] It should also be noted that the example embodiments mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in an order different from the embodiments, or several steps can be performed simultaneously.

[0149] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the computer readable medium having instructions stored therein comprises an article of manufacture including a computer program of instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks.

[0150] The above is merely specific implementation of the present application, and those skilled in the art can clearly understand the specific working process of the system, module and unit described above for the convenience and brevity of description, which can refer to the corresponding process in the foregoing method embodiments, and will not be described here. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed in the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A computing power network task scheduling method, used for scheduling a server, characterized in that, The method comprises the following steps: receiving a user's computing power service request; disassembling the original request task into a plurality of atomic tasks and constructing an atomic task sequence; segmenting the atomic task sequence into a plurality of time-sharing task sequences to be executed in sequence, and the execution period of each time-sharing task sequence coincides with an electricity price adjustment period; wherein the electricity price adjustment period is the time period from the end of the last electricity price adjustment to the beginning of the next electricity price adjustment; inputting the resource demand and task attribute information of each atomic task under the current time-sharing task sequence, the computing power state of each computing power node in the current global resource pool, the access grid electricity carbon emission factor and the access electricity price into a pre-trained matching model in time sequence to obtain a plurality of computing power nodes with the lowest electricity cost matched with each atomic task under the current time-sharing task sequence; scheduling each atomic task under the current time-sharing task sequence to the corresponding computing power node.

2. The method of claim 1, wherein, The segmentation of the atomic task sequence into a plurality of time-sharing task sequences to be executed in sequence comprises: estimating the total execution time of the atomic task sequence; calculating the first ratio of the total execution time to the electricity price adjustment period; rounding up the first ratio to obtain m; segmenting the atomic task sequence into m time-sharing task sequences to be executed in sequence; wherein the starting execution time of the i-th time-sharing task sequence is T0+(i-1)×T, the ending execution time is T0+i×T, m≥i≥1, T0 is the starting time of the electricity price adjustment period, and T is the length of the electricity price adjustment period.

3. The method of claim 1, wherein, The segmentation of the atomic task sequence into a plurality of time-sharing task sequences to be executed in sequence comprises: for each atomic task in the atomic task sequence, determining whether there is a serial atomic task to be executed in front; in the case that there is no serial atomic task to be executed in front, calculating the second ratio of the estimated execution time of the atomic task to the electricity price adjustment period; rounding up the second ratio to obtain n; in the case that n is equal to 1, the atomic task is not segmented, and the atomic task is only located in the first time-sharing task sequence; in the case that n is greater than 1, the atomic task is segmented, and the segmented atomic task is located in the first to the n-th time-sharing task sequence.

4. The method of claim 1, wherein, Before the receiving of the user's computing power service request, the method further comprises: receiving the computing grid resource registration information sent by the resource pool side server; labeling the computing power state, the access grid electricity carbon emission factor and the access electricity price of each computing power node in the resource pool according to the computing grid resource registration information; wherein the computing power state comprises a plurality of the following parameters: location, service level, electrical energy use efficiency, computing power, computing efficiency, computing power architecture and network bandwidth.

5. The method of claim 4, wherein, The method further comprises: in the case that the computing grid resource registration information has been updated, updating the computing power state, the access grid electricity carbon emission factor and the access electricity price of the computing power node in the currently labeled resource pool.

6. A task processing method for a resource pool side server, characterized in that, The method comprises the following steps: processing the computing power state information, the access grid electricity carbon emission factor and the access electricity price of the local resource pool to generate computing grid resource registration information; sending the computing grid resource registration information to a scheduling server; wherein the scheduling server is used to execute the computing power network task scheduling method of any one of claims 1-5. Receive the atomic task and computing power node information sent by the scheduling server; According to the atomic task and computing power node information, the computing network resource template is arranged for the atomic task; The computing network resource template is sent to the designated computing power node, and the computing power node is instantiated according to the computing network resource template, and the computing result is returned and the computing power node resource is released after the atomic task processing is completed or the running time arrives.

7. The method of claim 6, wherein, The method further comprises: Monitoring the computing power state information of the local resource pool, the access power grid power carbon emission factor and the access power price; In the case that any one of the computing power state information of the local resource pool, the access power grid power carbon emission factor and the access power price changes, updating the computing network resource registration information of the local resource pool; The updated computing network resource registration information is synchronized to the scheduling server.

8. A computing power network task scheduling apparatus, configured to schedule a server, characterized in that, Comprise: The receiving module is used for receiving the computing power service request of the user; The deconstruction module is used for decomposing the original request task into a plurality of atomic tasks and constructing an atomic task sequence; The segmentation module is used for segmenting the atomic task sequence into a plurality of time-sharing task sequences which are executed in sequence, and the execution period of each time-sharing task sequence coincides with an electricity price adjustment period; wherein the electricity price adjustment period is the time period between the end of the last electricity price adjustment and the beginning of the next electricity price adjustment; The matching module is used for inputting the resource demand and task attribute information of each atomic task under the current time-sharing task sequence, the computing power state of each computing power node in the current global resource pool, the access power grid power carbon emission factor and the access power price into a pre-trained matching model in time sequence, to obtain a plurality of computing power nodes with the lowest electricity cost matched with each atomic task under the current time-sharing task sequence; The scheduling module is used for scheduling each atomic task under the current time-sharing task sequence to the corresponding computing power node.

9. A task processing apparatus for a resource pool side server, characterized by comprising: Comprise: The generation module is used for processing the computing power state information of the local resource pool, the access power grid power carbon emission factor and the access power price to generate the computing network resource registration information; The first sending module is used for sending the computing network resource registration information to the scheduling server; wherein the scheduling server comprises the computing power network task scheduling device of claim 8; The receiving module is used for receiving the atomic task and computing power node information sent by the scheduling server; The arrangement module is used for arranging the computing network resource template for the atomic task according to the atomic task and computing power node information; The second sending module is used for sending the computing network resource template to the designated computing power node; The computing power node is used for instantiation according to the computing network resource template, and returns the computing result and releases the computing power node resource after the atomic task processing is completed or the running time arrives. 10.A computing power network task scheduling system, characterized in that, Comprise: The scheduling server, cloud, edge, end resource pool and a plurality of resource pool side servers deployed in the cloud, edge and end resource pool; The scheduling server comprises the computing power network task scheduling device of claim 8; and the resource pool side server comprises the task processing device of claim 9.

11. A terminal device, comprising: The device comprises a processor and a memory storing computer program instructions; The processor implements the computing power network task scheduling method according to any one of claims 1-5 or the task processing method according to claim 6 or 7 when executing the computer program instructions.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer program instructions, and the computer program instructions are executed by the processor to implement the computing power network task scheduling method according to any one of claims 1-5 or the task processing method according to claim 6 or 7.

Images