Hybrid computing power resource scheduling method and device

Abstract

The invention discloses a hybrid computing power resource scheduling method and device. The method comprises the following steps: acquiring a to-be-computed task information set and a computing resource information set; classifying the computing resource information set to obtain a computing resource subset corresponding to each task type; constructing a resource allocation model corresponding to each task type; performing resource allocation on the computing resource subsets corresponding to the task types by utilizing the resource allocation model corresponding to each task type to obtain a resource allocation matrix of the computing resource subsets; and scheduling and allocating all the computing resources by utilizing the resource allocation matrixes of all the computing resource subsets.

Description

Technical Field

[0001] The present invention relates to the fields of supercomputing technology, quantum computing and brain-inspired computing, and specifically to a resource scheduling method and device for hybrid computing power. Background Art

[0002] With the continuous advancement of computing technology, various computing architectures, such as GPUs (graphics processing units), CPUs (central processing units), quantum computing, and brain-inspired computing, have demonstrated their unique advantages in different application scenarios. However, effectively integrating and scheduling these different types of computing resources to achieve efficient computing task processing remains a pressing technical challenge.

[0003] Currently, common computing power scheduling methods typically target only a single type of computing architecture and fail to fully leverage the advantages of hybrid computing power. Furthermore, factors such as performance differences between different computing architectures, communication overhead, and task adaptability also pose significant challenges to hybrid computing power scheduling. Summary of the Invention

[0004] The present invention mainly solves the problem of how to effectively schedule and allocate resources for different types of computing architectures. The present invention discloses a resource scheduling method and device for hybrid computing power.

[0005] In a first aspect, an embodiment of the present invention discloses a resource scheduling method for hybrid computing power, comprising:

[0006] S1, obtain the information set of the task to be calculated and the information set of the computing resources;

[0007] S2, classifying the computing resource information set to obtain a computing resource subset corresponding to each task type;

[0008] S3, build the resource allocation model corresponding to each task type;

[0009] S4, using the resource allocation model corresponding to each task type, allocate resources to the computing resource subset corresponding to the task type to obtain the resource allocation matrix of the computing resource subset; using the resource allocation matrix of all computing resource subsets to realize the scheduling allocation of all computing resources.

[0010] The information set of tasks to be calculated includes the task type and requirement information set of each computing task; the requirement information set includes the computing scale value, data scale value, real-time requirement value, and power consumption requirement value; the computing resource information set includes the architecture type information and status information set of each computing resource; the status information set includes the available computing power scale value, load value, communication bandwidth value, available storage space value, data bus bandwidth, maximum available power value, heat dissipation capacity value, and current temperature value;

[0011] The task type includes real-time tasks, low-power tasks, and data-scale tasks;

[0012] The architecture type information includes GPU architecture, CPU architecture, quantum computing architecture, and brain-like computing architecture;

[0013] The resource allocation model corresponding to each task type is constructed, including:

[0014] For real-time tasks, low-power tasks and data-scale tasks, a real-time task resource allocation model, a low-power task resource allocation model and a data-scale task resource allocation model are constructed respectively.

[0015] The classifying process of the computing resource information set to obtain a computing resource subset corresponding to each task type includes:

[0016] S21, obtaining a preset standard state sequence and vector value range corresponding to all task types; the standard state sequence includes a standard value of available computing power scale, a standard value of load, a standard value of communication bandwidth, a standard value of available storage space, a standard value of parallel data bandwidth, a standard value of maximum available power, a standard value of heat dissipation capacity, and a standard value of current temperature;

[0017] S22, performing difference calculation processing on the state information set of each computing resource in the computing resource information set and the standard state sequence of all task categories, to obtain a difference feature vector corresponding to the computing resource;

[0018] S23, using the vector value interval corresponding to each task type, determining the vector value interval to which the difference feature vector corresponding to each computing resource belongs;

[0019] S24 , determining, according to the vector value interval determined for each computing resource, a task type corresponding to the vector value interval, which is the task type corresponding to the computing resource.

[0020] The expression for the difference calculation process is:

[0021]

[0022] Among them, cyt i is the i-th element of the difference feature vector corresponding to a computing resource, z j is the jth element of the state information set of the computing resource, rwp ij is the jth element of the standard state sequence of the i-th task type, and M is the total number of elements in the standard state sequence.

[0023] The expression of the real-time task resource allocation model is:

[0024]

[0025] ∑ i A i1 B1 ik ≥C1 k1 ,

[0026] ∑ i A i6 B1 ik ≤C1 k4 ,

[0027] Among them, C1() represents the real-time task resource allocation objective function, B1 represents the resource allocation matrix of the real-time task resource allocation model, and B1 ik The value of the element in the i-th row and k-th column of B1 is 1, which means that the i-th computing resource corresponding to the real-time task participates in the k-th real-time task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the real-time task does not participate in the k-th real-time task. i1 、A i2 、A i3 、A i4 、A i5 、A i6 They represent the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, and maximum available power standard value of the state information set of the i-th computing resource corresponding to the real-time task, respectively. k1 、C1 k2 、C1 k3 、C1 k4 They respectively represent the calculation scale value, data scale value, real-time requirement value, and power consumption requirement value in the requirement information set of the k-th real-time task.

[0028] The expression of the data scale task resource allocation model is:

[0029] maxC2(B2)=∑ i,k |A1 i4 A1 i5 B2 ik -C2 k2 |,

[0030]

[0031] ∑ i A1 i1 B2 ik ≥C2 k1 ,

[0032] ∑ i A1i6 B2 ik ≤C2 k4 ,

[0033] Among them, C2() represents the data scale task resource allocation objective function, B2 represents the resource allocation matrix of the data scale task resource allocation model, and B2 ik The value of the element in the i-th row and k-th column of is 1, which means that the i-th computing resource corresponding to the data scale task participates in the k-th data scale task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the data scale task does not participate in the k-th data scale task. i1 、A1 i2 、A1 i3 、A1 i4 、A1 i5 、A1 i6 C2 represents the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, and maximum available power standard value of the state information set of the i-th computing resource corresponding to the data scale task. k1 、C2 k2 、C2 k3 、C2 k4 They respectively represent the calculation scale value, data scale value, real-time requirement value, and power consumption requirement value in the requirement information set of the k-th data scale task.

[0034] The expression of the low-power task resource allocation model is:

[0035] minC3(B3)=∑ i,k [|A2 i6 B3 ik -C3 k4 |+A2 i7 B3 ik (f(C3 k5 )-

[0036] B3 ik A2 i8 )],

[0037]

[0038] ∑ i A2 i1 B3 ik ≥C3 k1 ,

[0039]

[0040] Among them, C3() represents the low-power task resource allocation objective function, B3 represents the resource allocation matrix of the low-power task resource allocation model, and B3 ik The value of the element in the i-th row and k-th column of is 1, which means that the i-th computing resource corresponding to the low-power task participates in the k-th low-power task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the low-power task does not participate in the k-th low-power task. i1 、A2 i2 、A2 i3 、A2 i4 、A2 i5 、A2 i6 、A2 i7 、A2 i8 C3 represents the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, maximum available power standard value, heat dissipation capacity value, and current temperature value of the state information set of the i-th computing resource corresponding to the low-power task. k1 、C3 k2 、C3 k3 、C3 k4 They respectively represent the computing scale value, data scale value, real-time requirement value, and power consumption requirement value in the demand information set of the k-th low-power task. f() is the power-temperature mapping function, which is constructed through the power consumption history data and temperature history data of the computing resources.

[0041] According to a second aspect of an embodiment of the present invention, a resource scheduling device for hybrid computing power is disclosed, the device comprising:

[0042] a memory storing executable program code;

[0043] a processor coupled to the memory;

[0044] The processor calls the executable program code stored in the memory to execute the resource scheduling method of hybrid computing power.

[0045] According to a third aspect of an embodiment of the present invention, a computer-storable medium is disclosed, wherein the computer-storable medium stores computer instructions. When the computer instructions are called by a computer, the computer instructions are used to execute the resource scheduling method for hybrid computing power.

[0046] According to a fourth aspect of an embodiment of the present invention, an information data processing terminal is disclosed, which is used to implement the resource scheduling method of hybrid computing power.

[0047] The beneficial effects of the present invention are:

[0048] The present invention can flexibly allocate and schedule computing resources according to the characteristics and requirements of computing tasks to adapt to different application scenarios; through real-time resource monitoring and adjustment, it ensures the efficient execution of tasks and reduces computing time and costs.

[0049] The hybrid computing power resource scheduling method provided by the embodiment of the present invention lays the foundation for accurate scheduling by obtaining a comprehensive and detailed set of information on tasks to be calculated and a set of computing resource information, covering task types and various types of demand information, computing resource architecture types and various state information. In the classification processing stage, the preset standard state sequence and vector value interval are used, combined with a unique difference calculation processing expression, to carefully classify the computing resources, which can accurately correspond each computing resource to the appropriate task type, effectively solving the problem of inaccurate matching of resources and tasks in traditional methods. By constructing a resource allocation model for each task type and allocating resources based on this, an accurate resource allocation matrix is obtained, and scientific scheduling of all computing resources is achieved. This method can significantly improve the rationality of resource allocation, ensure that real-time tasks meet real-time requirements, low-power tasks meet power consumption standards, and data-scale tasks are efficiently processed, greatly improving the overall operating efficiency and resource utilization of the hybrid computing power system, reducing energy consumption, and having high practicality and innovation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 4 is an implementation flow chart of the method of the present invention. DETAILED DESCRIPTION

[0051] In order to better understand the content of the present invention, an embodiment is given here.

[0052] Figure 1 4 is an implementation flow chart of the method of the present invention.

[0053] In a first aspect, an embodiment of the present invention discloses a resource scheduling method for hybrid computing power, comprising:

[0054] S1, obtaining a set of information about tasks to be calculated and a set of information about computing resources; the set of information about tasks to be calculated includes a task type and a set of requirement information for each computing task; the set of requirement information includes a computing scale value, a data scale value, a real-time requirement value, and a power consumption requirement value; the set of computing resource information includes an architecture type information and a set of status information for each computing resource; the set of status information includes an available computing power scale value, a load value, a communication bandwidth value, an available storage space value, a data bus bandwidth, a maximum available power value, a heat dissipation capacity value, and a current temperature value;

[0055] The task type includes real-time tasks, low-power tasks, and data-scale tasks;

[0056] The architecture type information includes GPU architecture, CPU architecture, quantum computing architecture, and brain-like computing architecture.

[0057] S2, classifying the computing resource information set to obtain a computing resource subset corresponding to each task type;

[0058] S3, build the resource allocation model corresponding to each task type;

[0059] S4, using the resource allocation model corresponding to each task type, allocate resources to the computing resource subset corresponding to the task type to obtain the resource allocation matrix of the computing resource subset; using the resource allocation matrix of all computing resource subsets to realize the scheduling allocation of all computing resources.

[0060] The classifying process of the computing resource information set to obtain a computing resource subset corresponding to each task type includes:

[0061] S21, obtaining a preset standard state sequence and vector value range corresponding to all task types; the standard state sequence includes a standard value of available computing power scale, a standard value of load, a standard value of communication bandwidth, a standard value of available storage space, a standard value of parallel data bandwidth, a standard value of maximum available power, a standard value of heat dissipation capacity, and a standard value of current temperature;

[0062] S22, performing difference calculation processing on the state information set of each computing resource in the computing resource information set and the standard state sequence of all task categories, to obtain a difference feature vector corresponding to the computing resource;

[0063] S23, using the vector value interval corresponding to each task type, determining the vector value interval to which the difference feature vector corresponding to each computing resource belongs;

[0064] S24 , determining, according to the vector value interval determined for each computing resource, a task type corresponding to the vector value interval, which is the task type corresponding to the computing resource.

[0065] The expression for the difference calculation process is:

[0066]

[0067] Among them, cyt i is the i-th element of the difference feature vector corresponding to a computing resource, z j is the jth element of the state information set of the computing resource, rwp ij is the jth element of the standard state sequence of the i-th task type, and M is the total number of elements in the standard state sequence;

[0068] The resource allocation model corresponding to each task type is constructed, including:

[0069] For real-time tasks, low-power tasks and data-scale tasks, a real-time task resource allocation model, a low-power task resource allocation model and a data-scale task resource allocation model are constructed respectively;

[0070] The expression of the real-time task resource allocation model is:

[0071]

[0072] ∑ i A i1 B1 ik ≥C1 k1 ,

[0073] ∑ i A i6 B1 ik ≤C1 k4 ,

[0074] Among them, C1() represents the real-time task resource allocation objective function, B1 represents the resource allocation matrix of the real-time task resource allocation model, and B1 ik The value of the element in the i-th row and k-th column of B1 is 1, which means that the i-th computing resource corresponding to the real-time task participates in the k-th real-time task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the real-time task does not participate in the k-th real-time task. i1 、A i2 、A i3 、A i4 、A i5 、A i6 They represent the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, and maximum available power standard value of the state information set of the i-th computing resource corresponding to the real-time task, respectively. k1 、C1 k2 、C1 k3 、C1 k4 They represent the computational scale value, data scale value, real-time requirement value, and power consumption requirement value in the requirement information set of the k-th real-time task respectively;

[0075] In traditional hybrid computing power resource scheduling, due to the lack of accurate quantitative matching of task requirements and resource status, real-time tasks often find it difficult to obtain resources that meet their computing scale, data scale, real-time requirements and power consumption requirements. The real-time task resource allocation model of the present invention clarifies whether each computing resource participates in a specific real-time task by clearly defining the resource allocation matrix B1. The objective function C1 (B1) is quantitatively calculated based on the computing power scale standard value, load standard value, communication bandwidth standard value, etc. of the computing resources corresponding to the real-time task and the computing scale value, real-time requirement value, etc. of the real-time task. It can accurately measure the degree of matching between resource allocation and task requirements, thereby achieving accurate adaptation of resources and effectively avoiding the problem that the real-time task cannot meet the real-time requirements due to improper resource allocation.

[0076] By maximizing the objective function C1 (B1), this solution can find the optimal resource allocation solution while satisfying various constraints. This optimization strategy can fully tap the potential of computing resources and improve the overall performance of the system. In traditional methods, due to the lack of an effective optimization mechanism, resource allocation is often based on experience or simple rules, which can easily lead to waste of resources and poor system performance. The real-time task resource allocation model of the present invention can find the most suitable computing resource combination for real-time tasks in a complex hybrid computing power environment through mathematical optimization methods, thereby improving the processing efficiency and quality of real-time tasks and further improving the overall operating efficiency of the hybrid computing power system.

[0077] The expression of the data scale task resource allocation model is:

[0078] maxC2(B2)=∑ i,k |A1 i4 A1 i5 B2 ik -C2 k2 |,

[0079]

[0080] ∑ i A1 i1 B2 ik ≥C2 k1 ,

[0081] ∑ i A 1i6 B2 ik ≤C2 k4 ,

[0082] Among them, C2() represents the data scale task resource allocation objective function, B2 represents the resource allocation matrix of the data scale task resource allocation model, and B2 ikThe value of the element in the i-th row and k-th column of is 1, which means that the i-th computing resource corresponding to the data scale task participates in the k-th data scale task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the data scale task does not participate in the k-th data scale task. i1 、A1 i2 、A1 i3 、A1 i4 、A1 i5 、A1 i6 C2 represents the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, and maximum available power standard value of the state information set of the i-th computing resource corresponding to the data scale task. k1 、C2 k2 、C2 k3 、C2 k4 They represent the computational scale value, data scale value, real-time requirement value, and power consumption requirement value in the requirement information set of the k-th data scale task respectively;

[0083] The expression of the low-power task resource allocation model is:

[0084] minC3(B3)=∑ i,k [|A2 i6 B3 ik -C3 k4 |+A2 i7 B3 ik (f(C3 k4 )-

[0085] B3 ik A2 i8 )],

[0086]

[0087] ∑ i A2 i1 B3 ik ≥C3 k1 ,

[0088]

[0089] Among them, C3() represents the low-power task resource allocation objective function, B3 represents the resource allocation matrix of the low-power task resource allocation model, and B3 ik The value of the element in the i-th row and k-th column of is 1, which means that the i-th computing resource corresponding to the low-power task participates in the k-th low-power task. ikThe value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the low-power task does not participate in the k-th low-power task. i1 、A2 i2 、A2 i3 、A2 i4 、A2 i5 、A2 i6 、A2 i7 、A2 i8 C3 represents the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, maximum available power standard value, heat dissipation capacity value, and current temperature value of the state information set of the i-th computing resource corresponding to the low-power task. k1 、C3 k2 、C3 k3 、C3 k4 They represent the computational scale value, data scale value, real-time requirement value, and power requirement value in the requirement information set of the k-th low-power task, respectively. f() is a power-temperature mapping function, which is constructed based on the historical power consumption and temperature data of the computing resources.

[0090] The first to eighth elements of the standard state sequence are, respectively, the standard value of available computing power scale, the standard value of load, the standard value of communication bandwidth, the standard value of available storage space, the standard value of parallel data bandwidth, the standard value of maximum available power, the standard value of heat dissipation capacity, and the standard value of current temperature;

[0091] The first to eighth elements of the state information set are respectively the available computing power scale value, load value, communication bandwidth value, available storage space value, data bus bandwidth, maximum available power value, heat dissipation capacity value, and current temperature value;

[0092] The resource allocation matrix of all computing resource subsets is used to achieve the scheduling and allocation of all computing resources. Each computing resource is allocated to a corresponding computing task according to the resource allocation matrix to achieve the scheduling and allocation of all computing resources.

[0093] The construction process of the power-temperature mapping function f() includes:

[0094] Obtaining a power consumption history data sequence set and a temperature history data sequence set of the computing resource; the power consumption history data sequence set includes a plurality of power consumption history data sequences; the temperature history data sequence set includes a plurality of temperature history data sequences;

[0095] The power consumption history data sequence set and the temperature history data sequence set are represented as a power consumption data matrix and a temperature data matrix respectively; the row vector of the power consumption data matrix is the power consumption history data sequence; the row vector of the temperature data matrix is the temperature history data sequence;

[0096] Based on the power consumption data matrix and the temperature data matrix, an optimized approximation model is constructed; the expression of the optimized approximation model is:

[0097] min|GA-T|,

[0098] subject to AA T =I A ,

[0099] Among them, I A represents the identity matrix with the row dimension of matrix A as the dimension, matrix A represents the optimization matrix to be solved, G represents the power consumption data matrix, and T represents the temperature data matrix;

[0100] Solving the optimization approximation model to obtain a matrix A;

[0101] Calculate the characteristic order of matrix A to obtain the characteristic order value;

[0102] The expression for calculating the characteristic order is:

[0103]

[0104] Among them, θ max Represents the maximum singular value of matrix A, T1 represents the rank value of matrix A, m is the row dimension of matrix A, a i is the mean of the i-th row of matrix A, v i is the maximum element value of the i-th singular vector of matrix A, Indicates rounding down.

[0105] Based on the characteristic order value, the power-temperature mapping function f() to be solved is constructed; the expression of f() is:

[0106]

[0107] Where N is the order of f(), α0, α1, α2,…, α N are the coefficients of each order of f(), x is the independent variable, and f(x) is the dependent variable corresponding to the independent variable;

[0108] Taking the power consumption historical data sequence as the known dependent variable and the temperature historical data sequence with the same collection time as the power consumption historical data sequence as the known independent variable, the power-temperature mapping function f() to be solved is fitted and solved using the known dependent variable and the known independent variable to obtain the power-temperature mapping function f().

[0109] The function fitting solution may adopt a polynomial function fitting method.

[0110] The optimization approximation model may be solved by using a numerical optimization method, a simplex method, or the like.

[0111] According to a second aspect of an embodiment of the present invention, a resource scheduling device for hybrid computing power is disclosed, the device comprising:

[0112] a memory storing executable program code;

[0113] a processor coupled to the memory;

[0114] The processor calls the executable program code stored in the memory to execute the resource scheduling method of hybrid computing power.

[0115] According to a third aspect of an embodiment of the present invention, a computer-storable medium is disclosed, wherein the computer-storable medium stores computer instructions. When the computer instructions are called by a computer, the computer instructions are used to execute the resource scheduling method for hybrid computing power.

[0116] According to a fourth aspect of an embodiment of the present invention, an information data processing terminal is disclosed, which is used to implement the resource scheduling method of hybrid computing power.

[0117] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A hybrid computing power resource scheduling method, characterized in that: include: S1, obtain the information set of the task to be calculated and the information set of the computing resources; S2, classifying the computing resource information set to obtain a computing resource subset corresponding to each task type; S3, build the resource allocation model corresponding to each task type; S4, using the resource allocation model corresponding to each task type, performing resource allocation on the computing resource subset corresponding to the task type, and obtaining a resource allocation matrix for the computing resource subset; The resource allocation matrix of all computing resource subsets is used to implement the scheduling and allocation of all computing resources.

2. The hybrid computing power resource scheduling method according to claim 1, characterized in that: The information set of tasks to be calculated includes the task type and requirement information set of each computing task; the requirement information set includes the computing scale value, data scale value, real-time requirement value, and power consumption requirement value; the computing resource information set includes the architecture type information and status information set of each computing resource; the status information set includes the available computing power scale value, load value, communication bandwidth value, available storage space value, data bus bandwidth, maximum available power value, heat dissipation capacity value, and current temperature value; The task type includes real-time tasks, low-power tasks, and data-scale tasks; The architecture type information includes GPU architecture, CPU architecture, quantum computing architecture, and brain-like computing architecture; The resource allocation model corresponding to each task type is constructed, including: For real-time tasks, low-power tasks and data-scale tasks, a real-time task resource allocation model, a low-power task resource allocation model and a data-scale task resource allocation model are constructed respectively.

3. The hybrid computing power resource scheduling method according to claim 2, characterized in that: The classifying process of the computing resource information set to obtain a computing resource subset corresponding to each task type includes: S21, obtaining a preset standard state sequence and vector value range corresponding to all task types; the standard state sequence includes a standard value of available computing power scale, a standard value of load, a standard value of communication bandwidth, a standard value of available storage space, a standard value of parallel data bandwidth, a standard value of maximum available power, a standard value of heat dissipation capacity, and a standard value of current temperature; S22, performing difference calculation processing on the state information set of each computing resource in the computing resource information set and the standard state sequence of all task categories, to obtain a difference feature vector corresponding to the computing resource; S23, using the vector value interval corresponding to each task type, determining the vector value interval to which the difference feature vector corresponding to each computing resource belongs; S24 , determining, according to the vector value interval determined for each computing resource, a task type corresponding to the vector value interval, which is the task type corresponding to the computing resource.

4. The hybrid computing power resource scheduling method according to claim 3, characterized in that: The expression for the difference calculation process is: Among them, cyt i is the i-th element of the difference feature vector corresponding to a computing resource, z j is the jth element of the state information set of the computing resource, rwp ij is the jth element of the standard state sequence of the i-th task type, and M is the total number of elements in the standard state sequence.

5. The hybrid computing power resource scheduling method according to claim 3, characterized in that: The expression of the real-time task resource allocation model is: Among them, C1() represents the real-time task resource allocation objective function, B1 represents the resource allocation matrix of the real-time task resource allocation model, and B1 ik The value of the element in the i-th row and k-th column of B1 is 1, which means that the i-th computing resource corresponding to the real-time task participates in the k-th real-time task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the real-time task does not participate in the k-th real-time task. i1 、A i2 、A i3 、A i4 、A i5 、A i6 They represent the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, and maximum available power standard value of the state information set of the i-th computing resource corresponding to the real-time task, respectively. k1 、C1 k2 、C1 k3 、C1 k4 They respectively represent the calculation scale value, data scale value, real-time requirement value, and power consumption requirement value in the requirement information set of the k-th real-time task.

6. The hybrid computing power resource scheduling method according to claim 3, characterized in that: The expression of the data scale task resource allocation model is: Among them, C2() represents the data scale task resource allocation objective function, B2 represents the resource allocation matrix of the data scale task resource allocation model, and B2 ik The value of the element in the i-th row and k-th column of is 1, which means that the i-th computing resource corresponding to the data scale task participates in the k-th data scale task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the data scale task does not participate in the k-th data scale task. i1 、A1 i2 、A1 i3 、A1 i4 、A1 i5 、A1 i6 C2 represents the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, and maximum available power standard value of the state information set of the i-th computing resource corresponding to the data scale task. k1 、C2 k2 、C2 k3 、C2 k4 They respectively represent the calculation scale value, data scale value, real-time requirement value, and power consumption requirement value in the requirement information set of the k-th data scale task.

7. The hybrid computing power resource scheduling method according to claim 3, characterized in that: The expression of the low-power task resource allocation model is: Among them, C3() represents the low-power task resource allocation objective function, B3 represents the resource allocation matrix of the low-power task resource allocation model, and B3 ik The value of the element in the i-th row and k-th column of is 1, which means that the i-th computing resource corresponding to the low-power task participates in the k-th low-power task. ik The value of the element in the i-th row and k-th column of is 0, which means that the i-th computing resource corresponding to the low-power task does not participate in the k-th low-power task. i1 、A2 i2 、A2 i3 、A2 i4 、A2 i5 、A2 i6 、A2 i7 、A2 i8 C3 represents the computing power scale standard value, load standard value, communication bandwidth standard value, available storage space standard value, parallel data standard bandwidth, maximum available power standard value, heat dissipation capacity value, and current temperature value of the state information set of the i-th computing resource corresponding to the low-power task. k1 、C3 k2 、C3 k3 、C3 k4 They respectively represent the computing scale value, data scale value, real-time requirement value, and power consumption requirement value in the demand information set of the k-th low-power task. f() is the power-temperature mapping function, which is constructed through the power consumption history data and temperature history data of the computing resources.

8. A hybrid computing power resource scheduling device, characterized in that: The device comprises: a memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the resource scheduling method for hybrid computing power according to any one of claims 1 to 7.

9. A computer storable medium, characterized in that The computer storable medium stores computer instructions, and when the computer instructions are called by a computer, they are used to execute the resource scheduling method for hybrid computing power according to any one of claims 1 to 7.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the resource scheduling method of the hybrid computing power according to any one of claims 1 to 7.

Images