Tidal power scheduling method and system, electronic device and storage medium

Abstract

The present disclosure provides a tidal computing power scheduling method, system, electronic device and storage medium to solve the problem of edge computing task exception risk caused by tidal computing power change. The method comprises: establishing an edge computing power resource pool, obtaining resource information of computing power nodes in each computing power domain, calculating the computing power value of the computing power nodes and saving them to the edge computing power resource pool; dividing the time-varying characteristics of the computing power nodes in the resource pool to determine the tidal computing power nodes in the resource pool and the corresponding tidal time period; receiving the business request information of the computing power demand side, calculating the required computing power size, and estimating the required time length to complete the computing power task; finding the computing power nodes with computing power value greater than the required computing power of the task and no overlap with the tidal time period in the time range from the computing power resource pool, and taking them as target computing power nodes. The present disclosure can avoid resource occupation peaks of computing power nodes when allocating computing power, and more reasonably match computing power demand.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of communication, in particular to a tidal computing power scheduling method, a tidal computing power scheduling system, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Cloud computing resources are often far away from terminals, so they cannot generally meet some applications with low delay requirements, which makes cloud computing resources have the characteristics of strong computing power, high energy consumption and high network delay. In comparison, edge computing devices are close to data sources, and are devices that integrate computing, storage, network and AI capabilities, and have the advantages of low latency, high bandwidth, local activity, local security and the like compared with cloud computing.

[0003] In modern industrial production environments, higher computing power and lower latency are often required, so edge computing has become the preferred solution. However, in existing edge computing power scheduling technologies, the matching degree of edge computing power is often limited, that is, only enough computing power is provided to ensure the computing tasks of demanders. This method ignores the risks caused by the fluctuation of potential available computing power of edge computing nodes, and the single-dimensional computing power resource scheduling method is not optimal in most cases.

[0004] From the perspective of edge computing power scheduling strategy, current edge computing users mainly develop scheduling systems through their own business characteristics to meet their business needs, and often use third-party provided general static scheduling strategies for scheduling. This static scheduling strategy does not take into account the dynamic changes of computing power demanders and providers, and in extreme cases may cause edge computing tasks to fail to be completed normally. SUMMARY

[0005] In order to at least solve the above technical problems existing in the prior art, the present disclosure provides a tidal computing power scheduling method, a tidal computing power scheduling system, an electronic device and a computer readable storage medium, which can avoid resource occupation peaks of computing power nodes when allocating computing power, more reasonably match computing power demand, improve the one-time completion rate of edge computing power tasks, avoid the problem that computing tasks fail due to fluctuations in available computing power resources of edge computing power nodes and then need to be allocated twice, and can better improve the efficiency of edge computing.

[0006] In a first aspect, the present disclosure provides a tidal computing power scheduling method, the method comprising:

[0007] establishing an edge computing power resource pool, obtaining resource information of computing power nodes in each computing power domain, calculating the computing power value of the computing power nodes according to the resource information, and saving the resource information and the computing power value of each computing power node to the edge computing power resource pool;

[0008] The computing power nodes in the edge computing power resource pool are divided according to time-varying characteristics to determine the tidal computing power nodes in the edge computing power resource pool and the tidal time periods corresponding to the tidal computing power nodes;

[0009] The service request information of the computing power demand side is received, the computing power required by the service request is calculated, and the time required to complete the computing power task is estimated;

[0010] From the computing power resource pool, the computing power nodes with computing power values greater than the required computing power of the task and without overlap with the tidal time period in the time range covering the task time are found as the target computing power nodes of the task.

[0011] Further,

[0012] The resource information includes device identification, CPU (Central Processing Unit) / GPU (graphics processing unit) frequency, core number, device memory and network bandwidth

[0013] The computing power value of the computing power node is calculated according to the resource information, which includes:

[0014] The computing power value of the computing power node is calculated according to the CPU / GPU parameters to obtain the quantized computing power value.

[0015] Further,

[0016] The computing power nodes in the edge computing power resource pool are divided according to time-varying characteristics, which includes:

[0017] The historical resource occupation of each computing power node in the edge computing power resource pool is counted one by one, and a first threshold is set. When the average resource occupation rate of a computing power node in a certain period is greater than the first threshold, it is identified as a tidal computing power node in the edge computing power resource pool, and the corresponding period is identified as the tidal time period of the tidal computing power node.

[0018] The method further includes:

[0019] After determining whether the computing power node is a tidal computing power node and the corresponding tidal time period, the characteristic information of all computing power nodes in the edge computing power resource pool is determined, including device identification, computing power value, whether it is a tidal computing power node and tidal time period.

[0020] Further, the method further includes:

[0021] The average resource occupation rate of the computing power node is calculated by the following formula:

[0022] Resource occupation rate = α*A + β*B + γ*C

[0023] Wherein, alpha, beta, gamma, respectively, are the CPU, memory, network bandwidth in the resource occupation calculation weight, A is the CPU occupation rate of the computing power node, B is the memory occupation rate of the computing power node, C is the network bandwidth occupation rate of the computing power node.

[0024] Further, the required computing power size of the service request is calculated, and the required duration for completing the computing power task is estimated, comprising:

[0025] The required computing power size of the service request is converted into FLOPS (Floating-point operations per second, floating-point operations per second);

[0026] According to the service request information of the computing power demander, it is classified into fixed calculation amount task and time length unestimable task;

[0027] For the fixed calculation amount task, the task duration is obtained by calculating the amount of calculation / power size, and for the time length unestimable task, the task duration is set to be unestimable.

[0028] Further, the method further comprises:

[0029] By analyzing the tidal computing power node associated with the tidal period T x [t i , t j ] is the pre-switching time period Delta t and the pre-switching time t i - Delta t of the tidal computing power node;

[0030] After reaching the pre-switching time, the task mirror of the task in the tidal computing power node is sent to the target computing power node for calculation.

[0031] Secondly, the present disclosure provides a tidal computing power scheduling system, comprising:

[0032] The resource pool module is configured to establish an edge computing power resource pool, obtain resource information of computing power nodes in each computing power domain, calculate the computing power value of the computing power nodes according to the resource information, and save the resource information and the computing power value of each computing power node to the edge computing power resource pool;

[0033] The division module is configured to perform time-varying characteristic division on the computing power nodes in the edge computing power resource pool to determine the tidal computing power nodes in the edge computing power resource pool and the tidal period corresponding to the tidal computing power nodes;

[0034] The demand analysis module is configured to receive service request information of the computing power demander, calculate the required computing power size of the service request, and estimate the required duration for completing the computing power task;

[0035] The matching module is configured to find, from the computing power resource pool, a computing power node with a computing power value greater than the required computing power of the task and with no overlap with the tidal period in the time range, and take the computing power node as a target computing power node of the task.

[0036] Further,

[0037] The dividing module is specifically configured to:

[0038] The historical resource occupation of each computing power node in the edge computing power resource pool is counted one by one, and a first threshold is set. When the average resource occupation rate of a computing power node in a period is greater than the first threshold, the computing power node is identified as a tidal computing power node in the edge computing power resource pool, and the corresponding period is identified as the tidal period of the tidal computing power node.

[0039] The resource pool module is further configured to:

[0040] After determining whether the computing power node is a tidal computing power node and the corresponding tidal period, the characteristic information of all computing power nodes in the edge computing power resource pool is determined, and the characteristic information includes device identification, computing power value, whether it is a tidal computing power node, and the tidal period.

[0041] In a third aspect, the present disclosure provides an electronic device including a memory and a processor, the memory storing a computer program, and when the processor executes the computer program stored in the memory, the processor executes the tidal computing power scheduling method according to any one of the first aspect.

[0042] In a fourth aspect, the present disclosure provides a computer readable storage medium, the computer readable storage medium storing a computer program, and when the computer program is executed by a processor, the tidal computing power scheduling method according to any one of the first aspect is implemented.

[0043] Beneficial effects:

[0044] The tidal computing power scheduling method, the tidal computing power scheduling system, the electronic device and the computer readable storage medium provided by the present disclosure integrate the computing power resources of heterogeneous computing power nodes, classify the edge computing power nodes according to their running rules through historical data statistical analysis, identify the time-varying rule of the available resources of the edge computing power nodes, avoid the resource occupation peak of the computing power nodes during computing power allocation, more reasonably match the computing power demand, improve the one-time completion rate of the edge computing power task, and avoid the problem that the computing task fails due to the fluctuation of the available computing power resources of the edge computing power nodes, and then needs to be allocated twice. The edge computing power resources can improve the efficiency of edge computing. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 It is a flowchart of a tidal computing power scheduling method provided by the first embodiment of the present disclosure.

[0046] Figure 2 A flowchart of a tidal computing power scheduling method provided for Embodiment Two of the present disclosure is shown in FIG. 1.

[0047] Figure 3 An architecture diagram of a tidal computing power scheduling system provided for Embodiment Three of the present disclosure is shown in FIG. 2.

[0048] Figure 4 An architecture diagram of an electronic device provided for Embodiment Four of the present disclosure is shown in FIG. 3. DETAILED DESCRIPTION

[0049] In order to enable a person skilled in the art to better understand the technical solutions of the present disclosure, the present disclosure will be described in further detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments and drawings described herein are merely for the purpose of explaining the present disclosure, and not limiting the present disclosure.

[0050] It should be noted that the terms “first”, “second”, etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence; and, in the case of no conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other at will.

[0051] The terms used in the embodiments of the present disclosure are merely for the purpose of describing specific embodiments, and are not intended to limit the present disclosure. The singular forms “a”, “an” and “the” used in the embodiments of the present disclosure and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0052] In the subsequent description, the suffixes such as “module”, “component”, or “unit” used to represent elements are merely for the purpose of facilitating the description of the present disclosure, and have no specific meaning in themselves. Therefore, “module”, “component”, or “unit” can be used mixedly.

[0053] For network edge computing resources, the emphasis is on making full use of the devices at each network edge to provide edge computing power services with moderate computing power, moderate energy consumption, and low network latency. With the rapid development of the Internet of Things, edge computing has gradually emerged. The emergence of edge computing has changed the relative independence of traditional clouds and networks, enabling computing to enter the network. The efficiency, credibility of edge computing will be deeply coupled with the bandwidth, latency, security, isolation of the network, and other aspects, to achieve efficient services of algorithm-network integration.

[0054] For edge computing users, how to make full use of the edge nodes they have involves the need for intelligent scheduling of edge nodes. Different users use different scheduling strategies according to different business scenarios. However, at present, edge computing users mainly develop scheduling systems according to their own business characteristics to meet their own business needs, and often use third-party provided general static scheduling strategies for scheduling. Such a static scheduling strategy does not take into account the dynamic changes of the computing power demand side and the provider, and in extreme cases, it may lead to the failure of normal completion of edge computing tasks.

[0055] The technical solutions of the present disclosure and how the technical solutions of the present disclosure solve the above technical problems in the prior art will be described in detail below with specific embodiments. It can be understood that in the embodiments of the present application, the execution subject can perform part or all of the steps in the embodiments of the present application, and these steps or operations are only examples, and the embodiments of the present application can also perform other operations or variations of various operations. In addition, each step can be executed in a different order as presented in the embodiments of the present application, and it is possible that not all operations in the embodiments of the present application are executed. In addition, the following specific embodiments can be combined with each other, and for the same or similar concepts or processes, they can not be described again in some embodiments.

[0056] Figure 1 A flowchart of a tidal computing power scheduling method provided by Embodiment One of the present disclosure is shown in FIG. 1, which includes the following steps. Figure 1

[0057] Step S101: Establish an edge computing power resource pool, obtain the resource information of the computing power nodes in each computing power domain, calculate the computing power value of the computing power nodes according to the resource information, and save the resource information and computing power value of each computing power node to the edge computing power resource pool;

[0058] Step S102: Perform time-varying feature division on the computing power nodes in the edge computing power resource pool to determine the tidal computing power nodes in the edge computing power resource pool and the tidal time period corresponding to the tidal computing power nodes.

[0059] Step S103: Receive the business request information of the computing power demand side, calculate the computing power size required by the business request, and estimate the time length required to complete the computing power task.

[0060] Step S104: Find a computing power node with a computing power value greater than the required computing power of the task and a time range covering the task time length without overlapping with the tidal time period from the computing power resource pool, and use it as the target computing power node for this task.

[0061] ​The edge computing task scheduling method provided by the embodiment of the present disclosure comprises the following steps: establishing an edge computing resource pool, collecting historical data of the use of computing resources of all edge computing nodes, and dividing the edge computing nodes into two categories: tidal computing nodes and non-tidal computing nodes; analyzing the details of the computing demand, including the business type, the required computing size, the expected task duration, etc., and then assigning the corresponding tidal computing nodes or non-tidal computing nodes to the computing demand, so as to effectively avoid the abnormal risk of edge computing tasks caused by the tidal computing change under the premise of fully meeting the computing demand of the task, and the problems of unreliable computing tasks, abnormal exit of edge computing tasks, and increase of the corresponding time delay caused by the re-distribution of edge computing tasks.

[0062] Further,

[0063] The resource information comprises device identification, CPU / GPU frequency, core number, device memory and network bandwidth

[0064] The computing of the computing value of the computing node according to the resource information comprises:

[0065] The computing value of the computing node is calculated according to the CPU / GPU parameter to obtain a quantized computing value.

[0066] The hardware resource information of all nodes (computing providers) in each computing domain is obtained, and after the quantized computing value is calculated according to the CPU / GPU parameter, the device identification, the computing value, the device memory, the network bandwidth and other information of each computing node are saved to the edge computing resource pool together;

[0067] In one embodiment, the computing value calculation method is: the computing value of the CPU / GPU chip = frequency * core number * floating point calculation value in a single cycle. For example, if the core number of a certain processor is 20, the frequency is 2.10 GHz, and the floating point calculation value in a single cycle is 32, then the computing value of the processor is: 20*2.10*32=1344 GFLOPS (FLOPS means floating point calculation times per second).

[0068] Further,

[0069] The time-varying feature division of the computing nodes in the edge computing resource pool comprises:

[0070] The historical resource occupation of each computing node in the edge computing resource pool is counted one by one, and a first threshold is set. When the average resource occupation rate of a certain computing node in a certain period is greater than the first threshold, the computing node is marked as a tidal computing node in the edge computing resource pool, and the corresponding period is marked as the tidal period of the tidal computing node.

[0071] The method further comprises:

[0072] Determine the characteristic information of all computing power nodes in the edge computing power resource pool after determining whether the computing power node is a tidal computing power node and the corresponding tidal period. The characteristic information includes device identification, computing power value size, whether it is a tidal computing power node, and a tidal period.

[0073] Considering that the resource utilization of edge computing power devices (including CPU, memory, available bandwidth, etc.) fluctuates, some edge computing power devices may have very high resource occupancy rate in a certain period or several periods. Such edge computing power device nodes are referred to as tidal computing power nodes, and vice versa. By adding the tidal computing power attribute to the edge computing power node, an effective reference can be provided for avoiding tidal periods.

[0074] In order to prevent the conflict between computing power scheduling and busy time of tidal computing power nodes from causing the computing power task to be unable to be completed normally, the historical occupancy of the resources of the edge computing power device is counted one by one, and a first threshold is set. When the average resource occupancy rate of a certain edge computing power device in a certain period is greater than the first threshold, it is identified as a tidal computing power node in the edge computing power resource pool, and the corresponding period T x [t i , t j ] is identified as a tidal period. An edge computing power device can have multiple tidal periods, which are represented by T = {T1, T2,... T x}. The first threshold can be set autonomously according to actual conditions, such as more than 70%, 80%, etc., which is not limited here.

[0075] Further, the method further comprises:

[0076] The average resource occupancy rate of the computing power node is calculated by the following formula:

[0077] Resource occupancy rate = a * A + b * B + g * C

[0078] Wherein, a, b, g are the weights of CPU, memory, and network bandwidth in resource occupancy calculation, A is the CPU occupancy rate of the computing power node, B is the memory occupancy rate of the computing power node, and C is the network bandwidth occupancy rate of the computing power node.

[0079] By setting the weights a, b, and g of the device CPU, memory, and network bandwidth in resource occupancy calculation, wherein a + b + g = 1, the average resource occupancy rate of the computing power node can be flexibly calculated, so that it is more suitable for the corresponding scene. For example, if the demand for computing tasks is biased towards CPU requirements, the weight of CPU in resource occupancy calculation can be increased.

[0080] Further, the size of the computing power required by the service request is calculated, and the time required to complete the computing power task is estimated, including:

[0081] The required computing power size of the service request is converted into FLOPS;

[0082] According to the service request information of the computing power demander, it is classified into fixed computing amount task and time length unestimable task;

[0083] For the fixed computing amount task, the task length is obtained by the computing amount / computing power size, and for the time length unestimable task, the task length is set to be unestimable.

[0084] After receiving the service request information from the computing power demander, the device identifier is obtained, the required computing power size is calculated, and the required computing power size is converted into FLOPS; according to the service request information of the computing power demander, it is classified into fixed computing amount task (such as AI training) and time length unestimable task (such as video monitoring); then the required time length for completing the computing power task is estimated: for the fixed computing amount task, the task length can be obtained by the computing amount / computing power size; for the time length unestimable task, the task length is also unestimable. According to the computing power demand value and the task length obtained in the foregoing steps, a matching node is searched in the edge computing power resource pool. For the computing task whose task length cannot be estimated, the tidal computing power node is avoided to be used, so as to avoid the risk that the computing task cannot be completed normally.

[0085] Further, the method further comprises:

[0086] By analyzing the tidal period T associated with the tidal computing power node x [t i , t j ] is set as the pre-switching time period Δt and the pre-switching time t i -Δt of the tidal computing power node;

[0087] After reaching the pre-switching time, the task mirror of the task in the tidal computing power node is sent to the target computing power node for calculation.

[0088] After obtaining the computing power demand value and the task length, the computing power value and the current system resource utilization of each computing power node in the computing power resource pool are obtained one by one, and the computing power node whose computing power value is greater than the required computing power of the task and whose task length covers the time range without overlapping with the tidal period T (which can be a tidal computing power node or a non-tidal computing power node) is searched from the computing power resource pool, and the computing power node is used as the target computing power node of the task;

[0089] And for the task in the tidal computing power node, through the pre-set pre-switching time period Δt (the length is set according to the volatility of the starting time of the tidal period) and the pre-switching time, the node that will enter the tidal period is prepared for switching by setting the pre-switching time and the target node, so as to avoid the risk of redundant calculation of the tidal node.

[0090] Further, the method further comprises:

[0091] The characteristic information of each computing power node in the edge computing power resource pool is updated regularly, so as to match the computing power demand through the updated edge computing power resource pool.

[0092] By updating the edge computing power resource pool, the characteristic information of each computing power node can be more accurate, thereby better realizing dynamic matching of computing power resources.

[0093] The edge computing power resource pool is updated, the characteristic information of each computing power node is more accurate, and the dynamic matching of computing power resources is better realized.

[0094] The second embodiment of the present disclosure also provides a tidal computing power scheduling method. By establishing an edge computing power resource pool, historical data of usage of computing power resources of all edge computing power nodes is statistically analyzed, and the edge computing power nodes are divided into two categories: tidal computing power nodes and non-tidal computing power nodes. A computing power matching module analyzes details of computing power demand, including a business type, a required computing power size, and a predicted task duration, and then allocates corresponding tidal computing power nodes or non-tidal computing power nodes to the computing power demand. Figure 2 As shown in the figure, the method specifically comprises:

[0095] Step S1: Establishing an edge computing power resource pool;

[0096] Hardware resource information of all nodes (computing power providers) in each computing power domain is acquired, including a device identifier, a CPU / GPU frequency, a core number, device memory, and network bandwidth. A node computing power value is calculated according to a CPU / GPU parameter, a quantized computing power value is obtained, and then the device identifier, the computing power value, the device memory, the network bandwidth, and the like are saved together to the edge computing power resource pool.

[0097] The computing power value calculation method is: the computing power of a CPU / GPU chip = frequency * core number * floating point calculation value in a single cycle. For example, if a processor has 20 cores, the frequency is 2.10 GHz, and the floating point calculation value in a single cycle is 32, then the computing power of the processor is: 20 * 2.10 * 32 = 1344 GFLOPS (FLOPS means floating point calculation times per second).

[0098] Step S2: Time-varying characteristic division of the edge computing power resource pool;

[0099] Considering that the resource utilization (including CPU, memory, available bandwidth, etc.) of the edge computing device fluctuates, some edge computing devices may have very high resource occupancy rate in a certain period or several periods. Such edge computing device nodes are referred to as tide nodes, and vice versa.

[0100] In order to prevent the conflict between the computing power scheduling and the busy time of the tide node from causing the computing power task to be unable to be completed normally, the historical occupancy of the edge computing device resource is counted one by one, and a first threshold is set. When the average resource occupancy rate of a certain edge computing device in a certain period is greater than the first threshold, it is identified as a tide computing node in the edge computing resource pool, and the corresponding period T x [t i , t j ] is identified as a tide period. There can be multiple tide periods for an edge computing device, which is represented by T = {T1, T2,... T x}.

[0101] The edge computing device resource occupancy rate calculation method: set three weight values, a, b, and g, which correspond to the weights of device CPU, memory, and network bandwidth in resource occupancy calculation. Set the CPU occupancy rate as A, the memory occupancy rate as B, and the network bandwidth occupancy rate as C. The resource occupancy rate of the corresponding device is: a*A + b*B + g*C.

[0102] Through the above method, the characteristic information of all computing power nodes in the edge computing resource pool is obtained, including device identification, computing power value, whether it is a tide node, and tide period T {T1, T2,... T x}.

[0103] Step S3: computing power demand analysis;

[0104] 1. Receive the service request information from the computing power demand side, obtain the device identification, and calculate the required computing power size, which is converted into FLOPS;

[0105] 2. Classify the computing power demand side according to the service request information, which is divided into fixed computing amount tasks (AI training, etc.) and time length unestimable tasks (video monitoring, etc.);

[0106] 3. Estimate the time length required to complete the computing power task: for fixed computing amount tasks, the task time length can be obtained by computing amount / computing power size; for time length unestimable tasks, the task time length is also unestimable.

[0107] 4. According to the computing power demand value and the task time length obtained in the foregoing steps, find the matching node in the edge computing resource pool.

[0108] Step S4: computing power matching;

[0109] 1. Obtain the computing power value of each computing power node in the computing power resource pool and the current system resource utilization (calculation method: step S2) ;

[0110] 2. Find a computing power node with a computing power value greater than the required computing power of the task and no overlap with the tidal period T in the time range covered by the task duration from the computing power resource pool (it can be a tidal node or a non-tidal node), and use it as the target computing power node for this task;

[0111] 3. Set a pre-switching time period Δt (the duration is set according to the volatility of the tidal period start time) and a pre-switching time t for the tidal node by analyzing the tidal period T associated with the tidal node i -Δt;

[0112] 4. After reaching the pre-switching time, send the task image to the target computing power node for calculation.

[0113] Through the above method, the edge computing task abnormal risk caused by tidal computing power changes can be effectively avoided under the premise of fully meeting the task computing power demand, and the problems of unreliable computing task, abnormal exit of edge computing task, and corresponding time delay increase caused by the reassignment of edge computing task can be solved.

[0114] Embodiment three of the present disclosure also provides a tidal computing power scheduling system, as shown in Figure 3 The system comprises:

[0115] A resource pool module 11 is configured to establish an edge computing power resource pool, obtain resource information of computing power nodes in each computing power domain, calculate the computing power value of the computing power nodes according to the resource information, and save the resource information and the computing power value of each computing power node to the edge computing power resource pool;

[0116] A division module 12 is configured to perform time-varying characteristic division on the computing power nodes in the edge computing power resource pool to determine the tidal computing power nodes in the edge computing power resource pool and the tidal periods corresponding to the tidal computing power nodes;

[0117] A demand analysis module 13 is configured to receive business request information from a computing power demand party, calculate the computing power required by the business request, and estimate the duration required to complete the computing power task;

[0118] A matching module 14 is configured to find a computing power node with a computing power value greater than the required computing power of the task and no overlap with the tidal period in the time range covered by the task duration from the computing power resource pool, and use it as the target computing power node for this task.

[0119] Further,

[0120] The resource information includes device identification, CPU / GPU frequency of the device, core number, device memory and network bandwidth

[0121] The resource pool module 11 is specifically configured as:

[0122] According to the CPU / GPU parameters, the computing power value of the computing power node is calculated to obtain a quantized computing power value.

[0123] Further,

[0124] The division module 12 is specifically configured as:

[0125] The historical resource occupation of each computing power node in the edge computing power resource pool is counted one by one, and a first threshold is set. When the average resource occupation rate of a computing power node in a certain period is greater than the first threshold, it is identified as a tidal computing power node in the edge computing power resource pool, and the corresponding period is identified as the tidal period of the tidal computing power node.

[0126] The resource pool module 11 is further configured as:

[0127] After determining whether the computing power node is a tidal computing power node and the corresponding tidal period, the feature information of all computing power nodes in the edge computing power resource pool is determined, including device identification, computing power value, whether it is a tidal computing power node and tidal period.

[0128] Further, the division module 12 is specifically further configured as:

[0129] The average resource occupation rate of the computing power node is calculated by the following formula:

[0130] Resource occupation rate = a*A + b*B + g*C

[0131] Wherein, a, b, g, are the weights of CPU, memory, network bandwidth in resource occupation calculation, A is the CPU occupation rate of the computing power node, B is the memory occupation rate of the computing power node, and C is the network bandwidth occupation rate of the computing power node.

[0132] Further, the demand analysis module 13 is specifically configured as:

[0133] The computing power required by the business request is converted into FLOPS;

[0134] According to the business request information of the computing power demand side, it is classified into fixed computing amount task and time length unestimable task;

[0135] For the fixed computing amount task, the task length is obtained by calculating the amount of calculation / power size, and for the time length unestimable task, the task length is set to be unestimable.

[0136] Furthermore, the system also includes a switching module 15;

[0137] The switching module 15 is configured to analyze the tidal period T associated with the tidal computing node. x [t i , t j Set the pre-switching time period Δt and pre-switching time t for the tidal computing power nodes. i -Δt; and,

[0138] After the pre-switch time is reached, the task image of the task in the tidal computing node is sent to the target computing node for calculation.

[0139] The tidal computing power scheduling system of this disclosure is used to implement the tidal computing power scheduling method in Method Embodiment 1 and Method Embodiment 2, so the description is relatively simple. For details, please refer to the relevant descriptions in Method Embodiment 1 and Method Embodiment 2 above, which will not be repeated here.

[0140] In addition, such as Figure 4 As shown, Embodiment 4 of this disclosure also provides an electronic device, including a memory 100 and a processor 200. The memory 100 stores a computer program. When the processor 200 runs the computer program stored in the memory 100, the processor 200 executes the various possible methods described above.

[0141] The memory 100 is connected to the processor 200. The memory 100 can be a flash memory, a read-only memory, or another type of memory. The processor 200 can be a central processing unit or a microcontroller.

[0142] Furthermore, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program, which is executed by a processor using the various possible methods described above.

[0143] The computer readable storage medium includes a volatile or non-volatile, removable or non-removable medium implemented in any method or technology for storage of information such as computer readable instructions, data structures, computer program modules or other data. The computer readable storage medium includes, but is not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile discs (DVD, Digital Video Disc) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer.

[0144] It can be understood that the above embodiments are only exemplary embodiments adopted for illustrating the principles of the present disclosure, and the present disclosure is not limited thereto. Various modifications and improvements can be made by those of ordinary skill in the art without departing from the spirit and essence of the present disclosure, and these modifications and improvements are also considered to be within the protection scope of the present disclosure.

Claims

1. A method for tidal computing power scheduling, characterized in that, The method comprises: Establishing an edge computing resource pool, obtaining resource information of each computing node in the computing domain, calculating the computing value of the computing node according to the resource information, and saving the resource information and the computing value of each computing node to the edge computing resource pool; The computing nodes in the edge computing resource pool are divided into time-varying characteristics to determine the tidal computing nodes in the edge computing resource pool and the tidal time period corresponding to the tidal computing nodes; After determining whether the computing node is a tidal computing node and the corresponding tidal time period, the characteristic information of all computing nodes in the edge computing resource pool is determined, including device identification, computing value size, whether it is a tidal computing node, and tidal time period; Receiving service request information of the computing demand side, calculating the computing size required by the service request, and estimating the time required to complete the computing task; From the computing resource pool, find the computing node whose computing value is greater than the required computing task, and the task time range does not overlap with the tidal time period, and use it as the target computing node of this task. Wherein, the time-varying characteristic division of the computing nodes in the edge computing resource pool comprises: The historical resource occupation of each computing node in the edge computing resource pool is counted one by one, and a first threshold is set. When the average resource occupation rate of a computing node in a certain period is greater than the first threshold, it is identified as a tidal computing node in the edge computing resource pool, and the corresponding period is identified as the tidal time period of the tidal computing node.

2. The method of claim 1, wherein The resource information includes device identification, device central processing unit (CPU) / graphics processing unit (GPU) frequency, core number, device memory and network bandwidth. The computing value of the computing node is calculated according to the CPU / GPU parameters to obtain the quantized computing value. The method further comprises:

3. The method of claim 1, wherein, The average resource occupation rate of the computing node is calculated by the following formula: Resource occupation rate = α * A + β * B + γ * C Wherein, α, β, γ, respectively, are the weights of CPU, memory and network bandwidth in resource occupation calculation, A is the CPU occupation rate of the computing node, B is the memory occupation rate of the computing node, and C is the network bandwidth occupation rate of the computing node. The computing size required by the service request is calculated, and the time required to complete the computing task is estimated, comprising:

4. The method of claim 1, wherein, Convert the computing size required by the service request into floating point operations per second (FLOPS); According to the service request information of the computing demand side, it is classified into fixed computing amount task and time length unestimable task; For fixed computing amount task, the task time is obtained by calculating the amount / computing size. For time length unestimable task, the task time is set to be unestimable. The method further comprises:

5. The method of claim 1, wherein, After reaching the pre-switching time, the task mirror of the task in the tidal computing node is sent to the target computing node for calculation. By analyzing the tidal period T associated with the tidal computing node x [t i , t j ] sets the pre-switching time period Δt and the pre-switching time t i -t; The system comprises:

6. A tidal hash power scheduling system, characterized in that, ​ The resource pool module is configured to establish an edge computing resource pool, acquire resource information of the computing nodes in each computing domain, calculate the computing values of the computing nodes according to the resource information, and save the resource information and the computing values of the computing nodes to the edge computing resource pool; The dividing module is configured to divide the computing nodes in the edge computing resource pool according to time-varying characteristics to determine the tidal computing nodes in the edge computing resource pool and the tidal time periods corresponding to the tidal computing nodes. The resource pool module is further configured to determine the characteristic information of all the computing nodes in the edge computing resource pool after determining whether the computing nodes are tidal computing nodes and the corresponding tidal time periods, and the characteristic information includes device identification, computing value size, whether a tidal computing node, and a tidal time period. The demand analysis module is configured to receive service request information of a computing demand party, calculate the computing size required by the service request, and estimate the time length required to complete the computing task. The matching module is configured to find, from the computing resource pool, a computing node whose computing value is greater than the computing required by the task and whose task time length does not overlap with the tidal time period in the time range, and take the computing node as a target computing node of the task. The dividing module is configured to: The historical resource occupation of each computing node in the edge computing resource pool is counted one by one, and a first threshold is set. When the average resource occupation rate of a computing node in a time period is greater than the first threshold, the computing node is identified as a tidal computing node in the edge computing resource pool, and the corresponding time period is identified as the tidal time period of the tidal computing node.

7. An electronic device, comprising: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the tidal computing scheduling method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the tidal computing scheduling method according to any one of claims 1-5.

Images