Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Collective communication function modeling method of parallel program

A technology of collective communication and modeling method, which is applied in the field of collective communication function modeling, can solve the problems of consuming a lot of time and money, and inaccurate data acquisition during communication, so as to save time and money and improve accuracy

Active Publication Date: 2016-12-21
HARBIN INST OF TECH
View PDF4 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to solve the disadvantages of inaccurate acquisition of communication time data and a large amount of time and money in the prior art, and propose a method for modeling collective communication functions of parallel programs

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 more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Collective communication function modeling method of parallel program
  • Collective communication function modeling method of parallel program
  • Collective communication function modeling method of parallel program

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0026] Specific implementation mode one: combine figure 1 To illustrate this embodiment, the specific process of a method for modeling collective communication functions of a parallel program in this embodiment is as follows:

[0027] Step 1, measure the collective communication function N times under the experimental platform, and obtain the communication time data of the collective communication function under different degrees of parallelism and data volume;

[0028] The value range of N is 1000-10000;

[0029] Step 2: Use the artificial neural network based on the BP (Error Back Propagation) backpropagation algorithm to fit the communication time data of the collective communication function under different degrees of parallelism and data volume, and obtain the neural network model of the corresponding communication function.

specific Embodiment approach 2

[0030] Specific embodiment two: the difference between this embodiment and specific embodiment one is: use the artificial neural network based on BP (Error Back Propagation) backpropagation algorithm in the described step 2 to set communication function in different degrees of parallelism and data volume The following communication time data are fitted to obtain the neural network model of the corresponding communication function; the specific process is:

[0031] The process of BP backpropagation algorithm is:

[0032] The forward propagation process first receives the input signal, and reaches the output layer through the weights and activation functions between neurons layer by layer to obtain the output value after the current iteration;

[0033] Calculate the error of this round of iteration according to the error definition method;

[0034] According to certain rules, the error is backpropagated from the output layer to the input layer, and the weight is adjusted layer ...

specific Embodiment approach 3

[0047] Specific implementation mode three: the difference between this implementation mode and specific implementation mode one or two is: the data preprocessing in the described step 21; the specific process is:

[0048] The communication time data under different parallelism and data volume is the communication time measured by changing the data volume under a certain parallelism degree or the communication time measured by changing the parallelism degree under a certain data volume;

[0049] The communication time data under different degrees of parallelism and data volume are aggregated and uneven, which has a great impact on training. Therefore, before transmitting the communication time data under different degrees of parallelism and data volume to network training, it is necessary to Shuffle the communication time data under different parallelism and data volume;

[0050]The parameter value update of the neurons in the first hidden layer is proportional to the input val...

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

No PUM Login to View More

Abstract

The invention relates to a collective communication function modeling method of a parallel program, and aims to eliminate the defects in the prior art that communication time data can not be accurately obtained and a great quantity of time and money is consumed. The collective communication function modeling method of the parallel program comprises the following specific process: S1: measuring a collective communication function under an experimental platform for N times, and obtaining the communication time data of the collective communication function under different degrees of parallelism and data sizes; and S2: utilizing an artificial neural network based on a BP (Back Propagation) algorithm to fit the communication time data of the collective communication function under different degrees of parallelism and data sizes to obtain a neural network model of a corresponding communication function. The collective communication function modeling method is used for the field of the communication technology.

Description

technical field [0001] The invention relates to a method for modeling collective communication functions of parallel programs. Background technique [0002] The execution time of a parallel program is divided into calculation and communication. The calculation time is the instruction execution time, and the communication time is the calling time of the communication function. In the research, the instruction execution time is obtained by dynamically counting the number of instructions and the execution time of various machine instructions, and the communication time is the focus of the research. The parallelism of scientific programs is usually implemented based on the MPI (Message-Passing Interface, MPI) interface, and MPI defines a function library that can be called by the programming language. The information of the communication function is obtained through instrumentation, the time model of the communication function is established, and the communication time is final...

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
no application Login to View More
IPC IPC(8): G06F19/00G06N3/02
CPCG06N3/02G16Z99/00
Inventor 张伟哲何慧郝萌韩硕鲁刚钊
Owner HARBIN INST OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com