Multi-thread application program dynamic scheduling method based on machine learning technology
A multi-threaded application and machine learning technology, applied in neural learning methods, multi-programming devices, program control design, etc., can solve the problem of unsatisfactory overall system performance improvement, long execution time of machine learning models, and inability to find optimal solutions for problems Solving problems such as solving problems, achieving the effects of reducing the number of times and online computing overhead, improving generalization performance, and improving overall performance
Inactive Publication Date: 2020-06-02
安徽安喆科技有限公司
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0005] The application of online scheduling has been proved to be an NP-complete problem, and the optimal solution of the problem cannot be found in polynomial time
Most of the existing solutions to NP-complete problems are based on heuristic algorithms or machine learning algorithms. Heuristic algorithms can only find suboptimal solutions, and the overall performance improvement effect of the system after scheduling is sometimes not ideal; traditional machine learning models take a long time to execute, and Frequent calls are required to find a relatively good solution, which brings huge scheduling overhead, which is even unacceptable in some cases
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0035] ANN mentioned in this embodiment: artificial neural network; BP: backpropagation; IPC: number of instructions executed per clock cycle.
[0036] Such as figure 2 As shown, in this embodiment, such as figure 1 As shown, a method for dynamically scheduling multi-threaded applications based on machine learning technology is applied to a multi-core execution platform with M different types of N processing cores. The multi-core execution platform used in the embodiment is as follows image 3 As shown, the processing core C1 is a large core (performance core), and the remaining three are small cores (energy efficiency cores). The specific process of assigning threads A, B, C, and D to the four processing cores C1~C4 is executed. And use the trained ANN to schedule threads A, B, C, D. Specifically, if figure 2 As shown, the scheduling method is carried out according to the following steps:
[0037] Step A. Initially, each thread in the multi-threaded application is rando...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a multi-thread application program dynamic scheduling method based on the machine learning technology, and the method comprises the steps: A) randomly mapping each thread in amulti-thread application program to each processing core in a multi-core execution platform at the beginning; B) collecting execution information of each thread in the multi-core execution platform atset intervals; C) detecting whether program behaviors are obviously changed or not according to the collected information: if so, performing remapping and scheduling, otherwise, repeating the step B); D) preprocessing the collected information to obtain performance prediction values of the thread on different processing cores; E) performing search evaluation on the mapping scheme from the threadto the processing core by utilizing the predicted value, and selecting the optimal mapping scheme to complete the mapping from the thread to the processing core. According to the invention, the optimal processing core is allocated to the program to reduce the calculation overhead of online scheduling, so the purpose of maximizing the overall performance of the system is achieved.
Description
technical field [0001] The invention relates to the fields of task scheduling and machine learning, in particular to a scheduling method for dynamically scheduling multi-threaded applications on a heterogeneous multi-core processing system based on machine learning technology to maximize performance. Background technique [0002] In order to meet the requirements of application scenarios for high performance and low power consumption, heterogeneous multi-core processing systems have gradually become the mainstream solution, and how to dynamically map and schedule resources for applications according to system requirements, so as to give full play to heterogeneous multi-core processing systems advantages, become an important problem that needs to be solved. [0003] The resource mapping and scheduling problem on the multi-core execution platform is essentially to allocate the application set to the processing core according to certain rules to give full play to the advantages...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06F9/50G06N3/08
CPCG06F9/5027G06F2209/5018G06F2209/508G06N3/084
Inventor 安鑫康安杨静卫圆祺王沐晗
Owner 安徽安喆科技有限公司