Heterogeneous computing power resource scheduling method and device based on user computing power demand, equipment and storage medium

Abstract

The invention discloses a heterogeneous computing power resource scheduling method and device based on user computing power demands, equipment and a storage medium, and the method comprises the steps: analyzing a task request, and obtaining a to-be-executed task and user resource demand information corresponding to the to-be-executed task; the user resource demand information comprises expected completion time and a maximum acceptable cost value; determining a target resource scheduling strategy based on the expected completion time and the maximum acceptable cost value; based on a preset constraint condition, selecting a plurality of computing nodes matched with the user resource demand information from all the computing nodes in the heterogeneous resource pool, and performing permutation and combination on the selected computing nodes to obtain a plurality of computing node combinations; selecting a target computing node combination from the plurality of computing node combinations based on the user resource demand information and a target resource scheduling strategy; the target computing node combination is used for executing the to-be-executed task. In this way, in the heterogeneous computing power resource scheduling process, the user satisfaction degree and the resource utilization rate are improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, apparatus, device and storage medium for scheduling heterogeneous computing resources based on user computing power requirements. Background Art

[0002] With the rapid development of cloud computing technology, heterogeneous computing resource pools (including various hardware types such as central processing units (CPUs), graphics processing units (GPUs), and field-programmable gate arrays (FPGAs)) have become an important infrastructure for supporting large-scale computing tasks.

[0003] Currently, research in the field of cloud computing task scheduling primarily uses scheduling methods based on intelligent optimization algorithms such as genetic algorithms and particle swarm optimization. However, these methods lack the ability to dynamically classify task strategies, making it impossible to flexibly adjust scheduling strategies, which in turn leads to low user satisfaction and resource utilization.

[0004] Therefore, in the process of scheduling heterogeneous computing resources, how to improve user satisfaction and resource utilization has become an urgent problem to be solved. Summary of the Invention

[0005] The embodiments of the present application provide a method, apparatus, device, and storage medium for scheduling heterogeneous computing resources based on user computing power requirements, which can improve user satisfaction and resource utilization during the process of scheduling heterogeneous computing resources.

[0006] In a first aspect, an embodiment of the present application provides a method for scheduling heterogeneous computing resources based on user computing power requirements, the method comprising:

[0007] Parse the task request from the client to obtain the task to be executed and the corresponding user resource demand information; the user resource demand information includes the expected completion time and maximum acceptable cost value of the task to be executed;

[0008] Determine the target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy;

[0009] Based on preset constraints, multiple computing nodes that match the user's resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations;

[0010] Based on user resource demand information and target resource scheduling policy, a target computing node combination is selected from multiple computing node combinations; the target computing node combination is used to execute the task to be executed.

[0011] In one embodiment, a target computing node combination is selected from a plurality of computing node combinations based on user resource demand information and a target resource scheduling policy, including: selecting at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from a plurality of computing node combinations based on the user resource demand information; determining a target weight corresponding to the target resource scheduling policy, and permuting and combining at least one first candidate computing node combination and at least one second candidate computing node combination to obtain a plurality of candidate computing node combinations; selecting a target computing node combination from a plurality of candidate computing node combinations based on the target weight, the total cost corresponding to the at least one first candidate computing node combination, and the completion time corresponding to the at least one second candidate computing node combination.

[0012] In one embodiment, based on user resource demand information, at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement are selected from multiple computing node combinations, including: determining the total cost of the task to be executed under each computing node combination, and selecting at least one computing node combination whose total cost is less than the maximum accepted cost value from the multiple computing node combinations corresponding to the total costs as at least one first candidate computing node combination; determining the completion time of the task to be executed under each computing node combination, and selecting at least one computing node combination whose completion time is prior to the expected completion time from the multiple computing node combinations corresponding to the completion times as at least one second candidate computing node combination.

[0013] In one embodiment, a target computing node combination is selected from multiple candidate computing node combinations based on the target weight, the total cost corresponding to at least one first candidate computing node combination, and the completion time corresponding to at least one second candidate computing node combination, including: inputting the target weight, expected completion time, maximum accepted cost value, and the total cost corresponding to the first candidate computing node combination and the completion time corresponding to the second candidate computing node combination in each candidate computing node combination into a pre-built matching score calculation model to obtain a matching score between each candidate computing node combination and the task to be executed; selecting the highest matching score from multiple matching scores, and using the candidate computing node combination corresponding to the highest matching score as the target computing node combination.

[0014] In one embodiment, a target resource scheduling strategy is determined based on the expected completion time and the maximum acceptance cost value, including: when the maximum acceptance cost value is less than a preset cost threshold and the expected completion time lags behind a first preset completion time, determining the target resource scheduling strategy as a cost priority strategy; when the expected completion time is prior to a second preset completion time, determining the target resource scheduling strategy as a time priority strategy; wherein the second preset completion time is prior to the first preset completion time.

[0015] In one embodiment, the method further includes: when any computing node in the target computing node combination is occupied and the preset resource preemption conditions are met, migrating the low-priority to-be-executed tasks originally assigned to the occupied computing node to other idle computing nodes for execution or delayed execution, and preempting the occupied computing node; wherein the preset resource preemption conditions include the following: the target resource scheduling policy is a time priority policy; the risk of the completion time of the to-be-executed tasks lagging behind the expected completion time is greater than a preset risk threshold; the completion time of the low-priority to-be-executed tasks is before the expected completion time.

[0016] In one embodiment, the method further includes: selecting, from a plurality of computing nodes other than the computing nodes included in the target computing node combination, computing nodes that are associated with the resource type required to execute the low-priority task to be executed and whose remaining storage space is greater than or equal to the data storage space required for the low-priority task to be executed, as other idle computing nodes.

[0017] In a second aspect, the present application provides a heterogeneous computing resource scheduling device based on user computing power requirements, the device comprising:

[0018] A parsing module is used to parse the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes the expected completion time and the maximum accepted cost value of the task to be executed;

[0019] A determination module is used to determine a target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy;

[0020] A selection module is used to select multiple computing nodes that match the user's resource demand information from all computing nodes in the heterogeneous resource pool based on preset constraints, and to arrange and combine the selected multiple computing nodes to obtain multiple computing node combinations;

[0021] The selection module is further used to select a target computing node combination from multiple computing node combinations based on user resource demand information and target resource scheduling strategy; the target computing node combination is used to execute the task to be executed.

[0022] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are performed:

[0023] Parse the task request from the client to obtain the task to be executed and the corresponding user resource demand information; the user resource demand information includes the expected completion time and maximum acceptable cost value of the task to be executed;

[0024] Determine the target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy;

[0025] Based on preset constraints, multiple computing nodes that match the user's resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations;

[0026] Based on user resource demand information and target resource scheduling policy, a target computing node combination is selected from multiple computing node combinations; the target computing node combination is used to execute the task to be executed.

[0027] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0028] Parse the task request from the client to obtain the task to be executed and the corresponding user resource demand information; the user resource demand information includes the expected completion time and maximum acceptable cost value of the task to be executed;

[0029] Determine the target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy;

[0030] Based on preset constraints, multiple computing nodes that match the user's resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations;

[0031] Based on user resource demand information and target resource scheduling policy, a target computing node combination is selected from multiple computing node combinations; the target computing node combination is used to execute the task to be executed.

[0032] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0033] Parse the task request from the client to obtain the task to be executed and the corresponding user resource demand information; the user resource demand information includes the expected completion time and maximum acceptable cost value of the task to be executed;

[0034] Determine the target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy;

[0035] Based on preset constraints, multiple computing nodes that match the user's resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations;

[0036] Based on user resource demand information and target resource scheduling policy, a target computing node combination is selected from multiple computing node combinations; the target computing node combination is used to execute the task to be executed.

[0037] The above-mentioned heterogeneous computing power resource scheduling method, device, equipment and storage medium based on user computing power demand, the computer equipment can parse the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes the expected completion time and maximum accepted cost value of the task to be executed; based on the expected completion time and maximum accepted cost value, the target resource scheduling strategy is determined; the target resource scheduling strategy is a cost priority strategy or a time priority strategy; based on preset constraints, multiple computing nodes matching the user resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations; based on the user resource demand information and the target resource scheduling strategy, a target computing node combination is selected from the multiple computing node combinations; the target computing node combination is used to execute the task to be executed. By adopting this method, since the user resource demand information is determined based on the user demand, during the process of heterogeneous computing power resource scheduling, the computer equipment can determine the target resource scheduling strategy that meets the user demand based on the user resource demand information. Then, based on the target resource scheduling strategy and the user resource demand information, the target computing node combination that meets the user demand and is used to execute the task to be executed can be determined from all the computing nodes in the heterogeneous resource pool, thereby improving user satisfaction and resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 This is a schematic diagram of an application scenario of a heterogeneous computing power resource scheduling method based on user computing power requirements provided by an embodiment of the present application;

[0040] Figure 2 This is a flow chart of a method for scheduling heterogeneous computing resources based on user computing power requirements, provided in an embodiment of the present application;

[0041] Figure 3 This is a flow chart of another method for scheduling heterogeneous computing resources based on user computing power requirements provided by an embodiment of the present application;

[0042] Figure 4 This is a structural diagram of a heterogeneous computing power resource scheduling device based on user computing power requirements provided by an embodiment of the present application;

[0043] Figure 5 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0045] The following introduces the application scenarios of the heterogeneous computing power resource scheduling method based on user computing power requirements provided in the embodiments of the present application.

[0046] See Figure 1 , Figure 1 This is a schematic diagram of an application scenario of a heterogeneous computing resource scheduling method based on user computing power requirements provided by an embodiment of the present application. Figure 1 As shown, the system includes a computer device 101 and a heterogeneous resource pool 102. Heterogeneous resource pool 102 includes multiple computing nodes, such as computing node 1021, computing node 1022, and computing node 1023. Each computing node may have the same or different resource types. Optionally, the resource types may include, but are not limited to, GPUs, CPUs, and FPGAs.

[0047] Among them, the computer device 101 can parse the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes time demand information and cost acceptance range; based on the time demand information and cost acceptance range, the target resource scheduling strategy is determined; the target resource scheduling strategy is a cost priority strategy or a time priority strategy; based on preset constraints, multiple computing nodes that match the user resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations; based on the user resource demand information and the target resource scheduling strategy, a target computing node combination is selected from the multiple computing node combinations; the target computing node combination is used to execute the task to be executed. By adopting this method, since the user resource demand information is determined based on the user demand, during the process of heterogeneous computing power resource scheduling, the computer equipment can determine the target resource scheduling strategy that meets the user demand based on the user resource demand information. Then, based on the target resource scheduling strategy and the user resource demand information, the target computing node combination that meets the user demand and is used to execute the task to be executed can be determined from all the computing nodes in the heterogeneous resource pool, thereby improving user satisfaction and resource utilization.

[0048] Optionally, the computer device 101 may be a server. The server mentioned here may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services, etc., which is not limited here.

[0049] See Figure 2 , Figure 2 This is a flow chart of a method for scheduling heterogeneous computing resources based on user computing power requirements provided by an embodiment of the present application. The method can be executed by a computer device (for example, the above-mentioned computer device 100). Figure 2 As shown, the heterogeneous computing resource scheduling method based on user computing power requirements may include but is not limited to the following steps:

[0050] S201: Parse the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes the expected completion time and the maximum accepted cost value of the task to be executed.

[0051] Optionally, the user resource requirement information also includes the submission time of the task to be executed, the amount of data corresponding to the task to be executed, the hardware model required to execute the task to be executed, the storage space, read / write performance, and upload / download bandwidth, etc., which are not limited here. The following uses Table 1 as an example to illustrate the user resource requirement information.

[0052] Table 1

[0053]

[0054] S202: Determine a target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy.

[0055] Optionally, the computer device may further determine a target resource scheduling policy based on other information included in the user resource requirement information. For example, assuming that the user resource requirement information includes the data volume of the task to be executed, the target resource scheduling policy determined by the computer device based on the data volume of the task to be executed may be a processing capacity priority policy. Assuming that the user resource requirement information includes the hardware model required to execute the task to be executed, the target resource scheduling policy determined by the computer device based on the hardware model required to execute the task to be executed may be a hardware model priority policy.

[0056] S203. Based on preset constraints, multiple computing nodes that match the user's resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations.

[0057] In an optional embodiment, the computer device selects multiple computing nodes that match the user's resource demand information from all computing nodes in the heterogeneous resource pool based on preset constraints, which may include: constructing a resource status matrix corresponding to the heterogeneous resource pool, the resource status matrix including status information of each computing node in the heterogeneous resource pool; based on the preset constraints, selecting multiple resource states that match the user's resource demand information from the resource status matrix, and using the computing nodes corresponding to the selected multiple resource states as the multiple computing nodes that match the user's resource demand information.

[0058] In some embodiments, the computer device constructs a resource status matrix corresponding to the heterogeneous resource pool, which may include but is not limited to the following steps:

[0059] Step 1: Obtain hardware attribute data of each computing node in the heterogeneous resource pool.

[0060] Optionally, the hardware attribute data of each computing node may include, but is not limited to, computing power (denoted as F_current), remaining memory capacity (denoted as STO_current), resource type (denoted as TYPE_current), read / write speed (denoted as IOPS_current), bandwidth (denoted as BANDWIDTH_current), and unit time cost (CUNIT_current). The resource type may be GPU, CPU, or FPGA, for example.

[0061] Step 2: Collect the occupancy time data of each computing node in the heterogeneous resource pool.

[0062] Optionally, the occupancy time related data includes the idle time window (denoted as W_idle). Where W_idle=[T_start,T_end], T_start represents the time when the computing node starts to be idle, and T_end represents the time when the computing node ends to be idle.

[0063] Step 3: Based on the hardware attribute related data and occupancy time related data of each computing node, the resource status matrix corresponding to the heterogeneous resource pool is generated.

[0064] Exemplarily, the resource status matrix corresponding to the heterogeneous resource pool is represented by the following expression (1).

[0065] (1)

[0066] In the above formula (1), R represents the resource status matrix corresponding to the heterogeneous resource pool; m represents the total number of computing nodes in the heterogeneous resource pool; r i1 It represents the computing capacity of the i-th computing node in the heterogeneous resource pool; r i2 It represents the resource type of the i-th computing node in the heterogeneous resource pool; r i3 It represents the remaining memory capacity of the i-th computing node in the heterogeneous resource pool; r i4 It represents the read and write rate of the i-th computing node in the heterogeneous resource pool; r i5 It represents the bandwidth of the i-th computing node in the heterogeneous resource pool; r i6 It represents the unit time cost of the i-th computing node in the heterogeneous resource pool; r i7 It represents the time when the i-th computing node in the heterogeneous resource pool starts to be idle; r i8 It represents the time when the i-th computing node in the heterogeneous resource pool ends being idle; where i is an integer greater than or equal to 1 and less than or equal to m.

[0067] In some embodiments, the preset constraints may include hardware type constraints, storage constraints, read / write capacity constraints, and bandwidth constraints. The hardware type constraint requires that the specific hardware model (or resource type) corresponding to the computing node matches the hardware model required to perform the task to be performed. The storage constraint requires that the remaining storage space of the computing node is greater than or equal to the storage space required to perform the task to be performed. The read / write capacity constraint requires that the read / write rate of the computing node is greater than or equal to the read / write performance required to perform the task to be performed. The bandwidth constraint requires that the upload / download bandwidth of the computing node is greater than or equal to the upload / download bandwidth required to perform the task to be performed.

[0068] The following is an example of how a computer device determines a combination of multiple computing nodes. For example, first, the computer device can select a resource type that is the same as TYPE_require in Table 1 from the resource status matrix shown in formula (1), that is, r i2 = TYPE_require, the remaining storage space is greater than or equal to STO_require in Table 1, that is, r i3 >STO_require elements, the read and write rate is greater than or equal to IOPS_require in Table 1, that is, r i4 >IOPS_require, and the bandwidth is greater than or equal to BANDWIDTH_require in Table 1, that is, r i5 >STO_require elements; then, the computer device may use the computing nodes corresponding to each selected element as multiple computing nodes that match the user resource requirement information. The computer device may then arrange and combine the multiple computing nodes that match the user resource requirement information to obtain multiple computing node combinations.

[0069] S204 : Based on the user resource demand information and the target resource scheduling policy, a target computing node combination is selected from the plurality of computing node combinations; the target computing node combination is used to execute the task to be executed.

[0070] In an optional implementation, after step S204 , the computer device may further assign the task to be executed to each computing node in the target computing node combination, so that each computing node in the target computing node combination executes the task to be executed.

[0071] In an embodiment of the present application, a computer device can parse a task request from a client to obtain a task to be executed and user resource demand information corresponding to the task to be executed; the user resource demand information includes time demand information and a cost acceptance range; based on the time demand information and the cost acceptance range, a target resource scheduling strategy is determined; the target resource scheduling strategy is a cost priority strategy or a time priority strategy; based on preset constraints, multiple computing nodes that match the user resource demand information are selected from all computing nodes in a heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations; based on the user resource demand information and the target resource scheduling strategy, a target computing node combination is selected from the multiple computing node combinations; the target computing node combination is used to execute the task to be executed. By adopting this method, since the user resource demand information is determined based on the user demand, during the process of heterogeneous computing power resource scheduling, the computer equipment can determine the target resource scheduling strategy that meets the user demand based on the user resource demand information. Then, based on the target resource scheduling strategy and the user resource demand information, the target computing node combination that meets the user demand and is used to execute the task to be executed can be determined from all the computing nodes in the heterogeneous resource pool, thereby improving user satisfaction and resource utilization.

[0072] In an optional embodiment, Figure 2 In the heterogeneous computing power resource scheduling method based on user computing power requirements shown, the computer device selects a target computing node combination from multiple computing node combinations based on user resource demand information and a target resource scheduling strategy, which may include: selecting at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from multiple computing node combinations based on the user resource demand information; determining the target weight corresponding to the target resource scheduling strategy, and arranging and combining at least one first candidate computing node combination and at least one second candidate computing node combination to obtain multiple candidate computing node combinations; selecting the target computing node combination from multiple candidate computing node combinations based on the target weight, the total cost corresponding to at least one first candidate computing node combination, and the completion time corresponding to at least one second candidate computing node combination.

[0073] In some embodiments, the computer device selects at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from multiple computing node combinations based on user resource demand information, which may include: determining the total cost of the task to be executed under each computing node combination, and selecting at least one computing node combination whose total cost is less than the maximum accepted cost value from the multiple computing node combinations corresponding to the total costs as at least one first candidate computing node combination; determining the completion time of the task to be executed under each computing node combination, and selecting at least one computing node combination whose completion time is prior to the expected completion time from the multiple computing node combinations corresponding to the completion times as at least one second candidate computing node combination.

[0074] Optionally, when determining the total cost of the task to be executed under each combination of computing nodes, the computer device may use the following formula (2).

[0075] (2)

[0076] In the above formula (2), It represents the total cost of any combination of computing nodes; It represents the unit time cost of the i-th computing node in the computing node combination; It represents the execution time of the task to be executed on the i-th computing node, where , F_task is the amount of data corresponding to the task to be executed, r i1 is the computing capacity of the i-th computing node; n represents the number of computing nodes in the computing node combination.

[0077] Optionally, the computer device selects at least one computing node combination whose total cost is less than the maximum accepted cost value from the computing node combinations corresponding to the multiple total costs as at least one first candidate computing node combination. as at least one first candidate computing node combination.

[0078] Optionally, when determining the completion time of the task to be executed under each combination of computing nodes, the computer device may use the following formula (3).

[0079] (3)

[0080] In the above formula (3), It represents the completion time of the task to be executed under any combination of computing nodes; It indicates the start time of the task to be executed in this computing node combination; It represents the execution time of the task to be executed in this computing node combination.

[0081] Optionally, the computer device selects at least one computing node combination whose completion time is earlier than the expected completion time from the computing node combinations corresponding to the multiple completion times as at least one second candidate computing node combination. as at least one second candidate computing node combination.

[0082] In some embodiments, the target weight corresponding to the target resource scheduling policy may be preset. For example, the target weight corresponding to the target resource scheduling policy is 0.8.

[0083] In some embodiments, based on the target weight, the total cost corresponding to at least one first candidate computing node combination and the completion time corresponding to at least one second candidate computing node combination, a target computing node combination is selected from multiple candidate computing node combinations, including: inputting the target weight, expected completion time, maximum accepted cost value, and the total cost corresponding to the first candidate computing node combination and the completion time corresponding to the second candidate computing node combination in each candidate computing node combination into a pre-built matching score calculation model to obtain a matching score between each candidate computing node combination and the task to be executed; selecting the highest matching score from multiple matching scores, and using the candidate computing node combination corresponding to the highest matching score as the target computing node combination.

[0084] Exemplarily, assuming that the number of the first candidate computing node combinations is 2 and the number of the second candidate computing node combinations is 2, the number of candidate node combinations is 2*2=4.

[0085] Optionally, a pre-built matching score calculation model may be as shown in the following formula (4).

[0086] (4)

[0087] In the above formula (4), It represents the matching score corresponding to the candidate computing node combination consisting of the i-th first candidate computing node combination and the j-th second candidate computing node combination; w cost It represents the weight of the cost priority strategy; COST_require represents the maximum accepted cost value; C total (i) represents the total cost of the i-th first candidate computing node combination; w timeIt represents the weight of the time priority strategy; TIME_require represents the expected completion time of the task to be executed; TIME_submit represents the submission time of the task to be executed; represents the completion time of the task to be executed under the jth second candidate computing node combination; where i and j are both positive integers greater than or equal to 1. It should be noted that w cost and w time The sum is 1. Optionally, when the target resource scheduling strategy is the cost priority strategy, w cost That is the target weight corresponding to the target resource scheduling strategy, w cost It can be 0.8. In this case, w time is 0.2; when the target resource scheduling strategy is the time priority strategy, w time That is the target weight corresponding to the target resource scheduling strategy, w time It can be 0.8. In this case, w cost is 0.2.

[0088] The following is an example of the process of determining the target computing node combination by a computer device.

[0089] For example, assuming there are two first candidate computing node combinations, denoted as combination 1 and combination 2, and two second candidate computing node combinations, denoted as combination 3 and combination 4, the computer device may determine that there are four candidate computing node combinations consisting of the two first candidate computing node combinations and the two second candidate computing node combinations, namely combination 1 and combination 3 (denoted as candidate computing node combination A), combination 1 and combination 4 (denoted as candidate computing node combination B), combination 2 and combination 3 (denoted as candidate computing node combination C), and combination 2 and combination 4 (denoted as candidate computing node combination D). If the matching scores corresponding to candidate computing node combination A, candidate computing node combination B, candidate computing node combination C, and candidate computing node combination D are 0.93, 0.95, 0.87, and 0.98, respectively, the computer device may determine that the highest matching score is 0.98. In this case, the computer device may select candidate computing node combination D, which corresponds to the highest matching score of 0.98, as the target computing node combination.

[0090] By adopting this implementation mode, the computer device can select a target computing node combination from multiple computing node combinations, and select a target computing node combination that meets the user resource demand information and the target resource scheduling strategy, thereby facilitating the assignment of tasks to be executed to the computing nodes in the target computing node combination for execution. In this way, not only can user satisfaction be improved, but resource utilization can also be improved.

[0091] In an optional embodiment, Figure 2In the heterogeneous computing power resource scheduling method based on user computing power requirements shown, the computer device determines the target resource scheduling strategy based on the expected completion time and the maximum acceptance cost value, which may include: when the maximum acceptance cost value is less than the preset cost threshold and the expected completion time lags behind the first preset completion time, determining the target resource scheduling strategy as a cost priority strategy; when the expected completion time is prior to the second preset completion time, determining the target resource scheduling strategy as a time priority strategy; wherein the second preset completion time is prior to the first preset completion time.

[0092] Optionally, the preset cost threshold can be 60% of the current average cost in the heterogeneous resource pool; the first preset completion time can be the current time + 24 hours (denoted as T_current + 24 hours). For example, assuming that the current time is 14:00 on April 29, 2025, the first preset completion time is 14:00 on April 30, 2025; the second preset completion time can be the current time + 6 hours (denoted as T_current + 6 hours). For example, assuming that the current time is 14:00 on April 29, 2025, the second preset completion time is 20:00 on April 29, 2025.

[0093] For example, assuming the maximum accepted cost is 1000, the preset cost threshold is 1500, the expected completion time is 16:00 on April 30, 2025, and the first preset completion time is 14:00 on April 30, 2025, the computer device can determine that the maximum accepted cost is less than the preset cost threshold and that the expected completion time lags behind the first preset completion time. In this case, the computer device can determine that the target resource scheduling policy is a cost-first policy. In this way, the computer device can prioritize scheduling low-cost computing nodes and allow pending tasks to be executed during resource-free periods (such as at night) to reduce the cost of executing pending tasks. For example, assuming the expected completion time is 16:00 on April 29, 2025, and the second preset completion time is 20:00 on April 29, 2025, the computer device can determine that the expected completion time takes precedence over the second preset completion time. In this case, the computer device can determine that the target resource scheduling policy is a time-first policy. In this way, the computer device can prioritize scheduling of high-performance computing nodes and immediately insert the tasks to be executed into the queue of tasks executed by the high-performance computing nodes to ensure that the tasks to be executed can be completed on time.

[0094] By adopting this implementation, the computer device determines the target resource scheduling strategy based on the expected completion time and the maximum acceptance cost value. In this way, by scheduling computing resources based on user needs, user satisfaction and resource utilization can be improved.

[0095] The following combination Figure 3, specifically describes the heterogeneous computing resource scheduling method based on user computing power requirements provided by the embodiment of this application. Figure 3 , Figure 3 This is a flow chart of another method for scheduling heterogeneous computing resources based on user computing power requirements provided by an embodiment of the present application. Figure 3 As shown, the heterogeneous computing resource scheduling method based on user computing power requirements may include but is not limited to the following steps:

[0096] S301: Parse the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes the expected completion time and the maximum accepted cost value of the task to be executed.

[0097] S302: Determine a target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy.

[0098] S303. Based on preset constraints, multiple computing nodes that match the user's resource demand information are selected from all computing nodes in the heterogeneous resource pool, and the selected multiple computing nodes are arranged and combined to obtain multiple computing node combinations.

[0099] In an optional implementation, the relevant descriptions of steps S301 to S303 can be found in the descriptions of the aforementioned steps S201 to S203, and will not be repeated here.

[0100] S304 : Based on the user resource requirement information, select at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from the multiple computing node combinations.

[0101] S305: Determine a target weight corresponding to the target resource scheduling strategy, and arrange and combine at least one first candidate computing node combination and at least one second candidate computing node combination to obtain a plurality of candidate computing node combinations.

[0102] S306. Based on the target weight, the total cost corresponding to at least one first candidate computing node combination, and the completion time corresponding to at least one second candidate computing node combination, select a target computing node combination from multiple candidate computing node combinations, and the target computing node is used to execute the task to be executed.

[0103] In an optional embodiment, the relevant description of steps S304 to S306 can be found in the aforementioned description of the computer device selecting at least one first candidate computing node combination that meets the cost requirements and at least one second candidate computing node combination that meets the time requirements from multiple computing node combinations based on user resource demand information, and will not be repeated here.

[0104] S307: Output the scheduling result including the target computing node combination.

[0105] Optionally, the scheduling results may include, but are not limited to, the target computing node combination for executing the pending task, the start time of the pending task, the estimated completion time of the pending task, and the total cost. This helps users clearly understand the execution plan corresponding to the pending task, and helps the system allocate computing resources based on the scheduling results and execute the pending task.

[0106] In an embodiment of the present application, since the user resource demand information is determined based on the user demand, during the process of scheduling heterogeneous computing resources, the computer device can determine the target resource scheduling strategy that meets the user demand based on the user resource demand information. Thereafter, based on the target resource scheduling strategy and the user resource demand information, a target computing node combination that meets the user demand and is used to execute the task to be executed can be determined from all the computing nodes in the heterogeneous resource pool, thereby improving user satisfaction and resource utilization.

[0107] In an optional embodiment, Figure 2 and Figure 3 In the heterogeneous computing resource scheduling method based on user computing power requirements, the computer device can also migrate low-priority pending tasks originally assigned to the occupied computing node to other idle computing nodes for execution or delay execution, and preempt the occupied computing node if any computing node in the target computing node combination is occupied and the preset resource preemption conditions are met. The preset resource preemption conditions include the following: the target resource scheduling policy is a time-priority policy; the risk of the completion time of the pending task lagging behind the expected completion time is greater than a preset risk threshold; and the completion time of the low-priority pending task is before the expected completion time. For example, the preset risk threshold is 80%.

[0108] That is to say, when multiple tasks compete for the same resource, for example, if any computing node in the target computing node combination used to execute the task to be executed is occupied by another task (for example, Task 1), if the corresponding target resource scheduling policy of the task to be executed is a time priority policy, and the risk that the completion time T_complete of the task to be executed exceeds its expected completion time TIME_require is higher than a preset threshold (such as 80%), and at the same time, the completion time T_complete of the preempted task (i.e., Task 1) is still before the expected completion time TIME_require of the task to be executed, then the computer device may migrate the low-priority task to be executed (Task 1) originally assigned to the occupied computing node to other idle computing nodes for execution or delay execution.

[0109] In some embodiments, other idle computing nodes can be determined by a computer device in the following manner: from multiple computing nodes other than the computing nodes included in the target computing node combination, computing nodes that are associated with the resource type required to execute low-priority tasks to be executed and whose remaining storage space is greater than or equal to the data storage space required for the low-priority tasks to be executed are selected as other idle computing nodes.

[0110] That is to say, in the process of migrating the preempted task to other idle computing nodes, the resource type of the other idle computing nodes must be the same as the resource type required to execute the preempted task, and the remaining storage space of the computing node must be greater than the data storage space required to execute the preempted task.

[0111] In some embodiments, the computer device may reinsert the preempted task into the scheduling queue if no other idle computing nodes are found, and then assign the preempted task to the occupied computing node when the occupied computing node is idle.

[0112] By adopting this embodiment, the computer device can dynamically adjust the task allocation scheme when any computing node in the target computing node combination is occupied, thereby ensuring the feasibility and efficiency of the resource scheduling result.

[0113] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0114] Based on the same inventive concept, the embodiments of the present application also provide a heterogeneous computing power resource scheduling device based on user computing power demand for implementing the above-mentioned heterogeneous computing power resource scheduling method based on user computing power demand. The implementation solution provided by this device is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations of one or more embodiments of the heterogeneous computing power resource scheduling device based on user computing power demand provided below can be found in the above-mentioned limitations on the heterogeneous computing power resource scheduling method based on user computing power demand, and will not be repeated here.

[0115] See Figure 4 , Figure 4 This is a schematic diagram of the structure of a heterogeneous computing resource scheduling device based on user computing power requirements provided by an embodiment of the present application. Figure 4 As shown, the heterogeneous computing resource scheduling device based on user computing power requirements may include but is not limited to:

[0116] The parsing module 401 is used to parse the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes the expected completion time and the maximum accepted cost value of the task to be executed;

[0117] Determination module 402, for determining a target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy;

[0118] The selection module 403 is configured to select, from all computing nodes in the heterogeneous resource pool, a plurality of computing nodes that match the user's resource demand information based on preset constraints, and to arrange and combine the selected plurality of computing nodes to obtain a plurality of computing node combinations;

[0119] The selection module 403 is further configured to select a target computing node combination from multiple computing node combinations based on user resource demand information and a target resource scheduling policy; the target computing node combination is used to execute the task to be executed.

[0120] In one embodiment, when the selection module 403 is used to select a target computing node combination from multiple computing node combinations based on user resource demand information and a target resource scheduling policy, it is specifically used to: select at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from multiple computing node combinations based on the user resource demand information; determine the target weight corresponding to the target resource scheduling policy, and arrange and combine at least one first candidate computing node combination and at least one second candidate computing node combination to obtain multiple candidate computing node combinations; select the target computing node combination from multiple candidate computing node combinations based on the target weight, the total cost corresponding to at least one first candidate computing node combination, and the completion time corresponding to at least one second candidate computing node combination.

[0121] In one embodiment, when the selection module 403 is used to select at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from multiple computing node combinations based on user resource demand information, it is specifically used to: determine the total cost of the task to be executed under each computing node combination, and select at least one computing node combination whose total cost is less than the maximum accepted cost value from the multiple computing node combinations corresponding to the total costs as at least one first candidate computing node combination; determine the completion time of the task to be executed under each computing node combination, and select at least one computing node combination whose completion time is prior to the expected completion time from the multiple computing node combinations corresponding to the completion times as at least one second candidate computing node combination.

[0122] In one embodiment, when the selection module 403 is used to select a target computing node combination from multiple candidate computing node combinations based on the target weight, the total cost corresponding to at least one first candidate computing node combination, and the completion time corresponding to at least one second candidate computing node combination, the selection module 403 is specifically used to: input the target weight, expected completion time, maximum accepted cost value, and the total cost corresponding to the first candidate computing node combination and the completion time corresponding to the second candidate computing node combination in each candidate computing node combination into a pre-built matching score calculation model to obtain a matching score between each candidate computing node combination and the task to be executed; select the highest matching score from multiple matching scores, and use the candidate computing node combination corresponding to the highest matching score as the target computing node combination.

[0123] In one embodiment, when the determination module 402 is used to determine the target resource scheduling strategy based on the expected completion time and the maximum acceptance cost value, it is specifically used to: when the maximum acceptance cost value is less than a preset cost threshold and the expected completion time lags behind the first preset completion time, determine the target resource scheduling strategy as a cost priority strategy; when the expected completion time is prior to the second preset completion time, determine the target resource scheduling strategy as a time priority strategy; wherein the second preset completion time is prior to the first preset completion time.

[0124] In one embodiment, the device may also include a processing module, which is used to migrate the low-priority to-be-executed tasks originally assigned to the occupied computing node to other idle computing nodes for execution or delayed execution, and preempt the occupied computing node when any computing node in the target computing node combination is occupied and the preset resource preemption conditions are met; wherein the preset resource preemption conditions include the following: the target resource scheduling policy is a time priority policy; the risk of the completion time of the to-be-executed task lags behind the expected completion time is greater than the preset risk threshold; the completion time of the low-priority to-be-executed task is before the expected completion time.

[0125] In one embodiment, the selection module 403 is also used to select, from multiple computing nodes other than the computing nodes included in the target computing node combination, computing nodes that are associated with the resource type required to execute the low-priority task to be executed and whose remaining storage space is greater than or equal to the data storage space required for the low-priority task to be executed, as other idle computing nodes.

[0126] Each module in the above-mentioned heterogeneous computing resource scheduling device based on user computing power requirements can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above modules can be embedded in or independent of the processor in the terminal device in hardware form, or can be stored in the memory of the terminal device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0127] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, memory and input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store a pre-built heterogeneous computing power resource pool. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a heterogeneous computing power resource scheduling method based on user computing power requirements is implemented.

[0128] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0129] In an exemplary embodiment, the present application provides a computer device including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the above-mentioned method for scheduling heterogeneous computing resources based on user computing power requirements are implemented.

[0130] In an exemplary embodiment, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-mentioned method for scheduling heterogeneous computing resources based on user computing power requirements.

[0131] In an exemplary embodiment, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps in the above-mentioned method for scheduling heterogeneous computing resources based on user computing power requirements.

[0132] It should be noted that the data involved in this application (including but not limited to task requests, user resource demand information, target resource scheduling strategies, target computing node combinations, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0133] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0134] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0135] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for scheduling heterogeneous computing resources based on user computing power requirements, characterized in that: The method comprises: Parsing the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes the expected completion time and the maximum accepted cost value of the task to be executed; Determine a target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy; Based on preset constraints, selecting a plurality of computing nodes that match the user resource demand information from all computing nodes in the heterogeneous resource pool, and permuting and combining the selected plurality of computing nodes to obtain a plurality of computing node combinations; Based on the user resource demand information and the target resource scheduling policy, a target computing node combination is selected from the plurality of computing node combinations; the target computing node combination is used to execute the task to be executed.

2. The method according to claim 1, characterized in that The selecting a target computing node combination from the plurality of computing node combinations based on the user resource demand information and the target resource scheduling policy includes: Based on the user resource demand information, selecting at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from the plurality of computing node combinations; Determining a target weight corresponding to the target resource scheduling strategy, and permuting and combining at least one of the first candidate computing node combinations and at least one of the second candidate computing node combinations to obtain a plurality of candidate computing node combinations; A target computing node combination is selected from the plurality of candidate computing node combinations based on the target weight, the total cost corresponding to at least one of the first candidate computing node combinations, and the completion time corresponding to at least one of the second candidate computing node combinations.

3. The method according to claim 2, characterized in that The selecting, based on the user resource demand information, at least one first candidate computing node combination that meets the cost requirement and at least one second candidate computing node combination that meets the time requirement from the plurality of computing node combinations comprises: Determining a total cost of the task to be executed under each of the computing node combinations, and selecting, from the computing node combinations corresponding to the multiple total costs, at least one computing node combination whose total cost is less than the maximum accepted cost value as at least one first candidate computing node combination; Determine the completion time of the task to be executed under each of the computing node combinations, and select at least one of the computing node combinations whose completion time is prior to the expected completion time from the computing node combinations corresponding to the multiple completion times as at least one second candidate computing node combination.

4. The method according to claim 3, characterized in that The selecting a target computing node combination from the plurality of candidate computing node combinations based on the target weight, the total cost corresponding to at least one of the first candidate computing node combinations, and the completion time corresponding to at least one of the second candidate computing node combinations includes: Inputting the target weight, the expected completion time, the maximum accepted cost value, and the total cost corresponding to the first candidate computing node combination and the completion time corresponding to the second candidate computing node combination in each candidate computing node combination into a pre-built matching score calculation model to obtain a matching score between each candidate computing node combination and the task to be executed; A highest matching score is selected from the plurality of matching scores, and the candidate computing node combination corresponding to the highest matching score is used as the target computing node combination.

5. The method according to claim 1, wherein The determining of the target resource scheduling strategy based on the expected completion time and the maximum accepted cost value includes: When the maximum accepted cost value is less than a preset cost threshold and the expected completion time lags behind a first preset completion time, determining the target resource scheduling strategy to be a cost priority strategy; In a case where the expected completion time takes precedence over a second preset completion time, the target resource scheduling policy is determined to be a time priority policy; wherein the second preset completion time takes precedence over the first preset completion time.

6. The method according to any one of claims 1 to 5, characterized in that The method further comprises: If any computing node in the target computing node combination is occupied and the preset resource preemption condition is met, the low-priority to-be-executed tasks originally assigned to the occupied computing node are migrated to other idle computing nodes for execution or delayed execution, and the occupied computing node is preempted; The preset resource preemption conditions include the following: The target resource scheduling strategy is a time priority strategy; The risk that the completion time of executing the task to be executed lags behind the expected completion time is greater than a preset risk threshold; The completion time of the low-priority task to be executed is before the expected completion time.

7. The method according to claim 6, characterized in that The method further comprises: From multiple computing nodes other than the computing nodes included in the target computing node combination, select computing nodes that are associated with the resource type required to execute the low-priority task to be executed and have remaining storage space greater than or equal to the data storage space required for the low-priority task to be executed as the other idle computing nodes.

8. A heterogeneous computing resource scheduling device based on user computing power requirements, characterized in that: The device comprises: A parsing module is used to parse the task request from the client to obtain the task to be executed and the user resource demand information corresponding to the task to be executed; the user resource demand information includes the expected completion time and the maximum accepted cost value of the task to be executed; A determination module is configured to determine a target resource scheduling strategy based on the expected completion time and the maximum accepted cost value; the target resource scheduling strategy is a cost priority strategy or a time priority strategy; A selection module is configured to select, from all computing nodes in the heterogeneous resource pool based on preset constraints, a plurality of computing nodes that match the user resource demand information, and to arrange and combine the selected plurality of computing nodes to obtain a plurality of computing node combinations; The selection module is further configured to select a target computing node combination from the plurality of computing node combinations based on the user resource demand information and the target resource scheduling policy; the target computing node combination is configured to execute the task to be executed.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images