Unlock instant, AI-driven research and patent intelligence for your innovation.

Space computing parallel load balancing method

A load balancing and space computing technology, applied in the field of high-performance computing, can solve the problems of increasing the area of ​​the divided surface, reducing the amount of communication, and increasing the communication overhead, so as to reduce the number of particles and reduce the communication load.

Active Publication Date: 2014-04-02
TONGJI UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Some researchers consider dynamic balance, such as using small grid units to divide, and when the load is unbalanced, the small grid units are completely migrated. This division method increases the area of ​​​​the division surface, thereby increasing the communication overhead.
Some use convex quadrilaterals for division, and use grid evolution for dynamic load balancing, but also ignore the requirement that the division surface should avoid particle-intensive areas, and the perimeter of the convex quadrilateral is still too large for a certain area. long, which is still not conducive to reducing traffic

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
  • Space computing parallel load balancing method
  • Space computing parallel load balancing method
  • Space computing parallel load balancing method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0064] see figure 1 , the present invention discloses a space computing parallel load balancing method, the method comprising:

[0065] [Step S1] Set the number of processors to be N, and generate corresponding N bubbles with the same or different sizes. For all bubbles on a two-dimensional plane, they are expressed as N disjoint circles; the number of particles in the simulation space for W, is the average value of the overall load, and the number of particles in each bubble is W 1 , W 2 ...W N-1 , W N .

[0066] A k×k sliding window is used to detect the density distribution in the simulation space, and the top N density extreme points with the highest density are selected as the starting positions of the bubbles.

[0067] [Step S2] Calculate the number of particles in each bubble, compare with the average value to determine expansion and contraction, and adjust the size of the bubble; when the number of internal particles is less than the average value, it will expan...

Embodiment 2

[0089] First, assuming that the number of processors is N, corresponding N bubbles are generated, and the sizes of the bubbles can be the same or different. For all bubbles on a two-dimensional plane, they are expressed as N disjoint circles. Assuming that the number of particles in the simulation space (which can also be regarded as the calculation load) is W, is the average value of the overall load, and the number of particles in each bubble is W 1 , W 2 ...W N-1 , W N , then the load balancing condition is: And W 1 =W 2 =...W N . The construction process of this model is completely formed by the expansion and mutual extrusion of the bubbles. The various behaviors of the bubbles and the formation algorithm of the model are described below:

[0090] Expansion and contraction of bubbles

[0091] The expansion and contraction of the bubble depend on the number of internal particles. When the number of internal particles is less than the average, it will expand, othe...

Embodiment 3

[0116] In this embodiment, approximately 11,000 particles are unevenly generated in a two-dimensional space of 800×500 pixels, and a 5×5 sliding window is used to determine the dense point of particle distribution. In the algorithm, the value of β is 1, and the value of δ is 0.1. Finally, The result is as image 3 shown. In addition, we use the same particle distribution to divide the space based on the quadrilateral mesh model. The final result is as follows Figure 4 shown.

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 discloses a space computing parallel load balancing method. The method comprises the following steps of: S1, detecting the density distribution condition in a simulation space by using a sliding window, and selecting front N density extreme points with highest density as initial positions of bubbles; S2, acquiring particle number in each bubble, comparing the particle number with an average value to determine expansion and contraction, and regulating the size of the bubbles, wherein when the internal particle number is smaller than the average value, the bubbles are expanded, otherwise, the bubbles are contracted; S3, acquiring a resultant force of a gradient force and an extrusion force, and regulating the positions of the bubbles; and S4, judging whether the particles in the simulation space form a division, if so, completing, otherwise, returning to the S2. The space computing parallel load balancing method provided by the invention reduce the particle number on a boundary by avoidance of the bubbles to a particle congestion area so as to reduce the communication load in a computing process.

Description

technical field [0001] The invention belongs to the technical field of high-performance computing and relates to a parallel load balancing method, in particular to a task division and load balancing method for parallel space computing. Background technique [0002] There is such a type of calculation in scientific computing, from microscopic molecular motion simulation, cell group growth simulation to macroscopic large-scale traffic simulation and even celestial body deduction. The objects of this type of simulation can be expressed as a large amount of interacting particles in space. This type of simulation usually involves large computing and storage requirements, and high-performance clusters are often required for distributed simulation. In this kind of distributed simulation, how to divide the simulation task is the key to realize the efficiency of parallel computing. In addition, another characteristic of this type of simulation is the fluidity of simulation objects, ...

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
Patent Type & Authority Patents(China)
IPC IPC(8): G06F9/46
Inventor 蒋昌俊张栋良陈闳中闫春钢丁志军张亚英
Owner TONGJI UNIV
Features
  • R&D
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More

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

© 2025 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com