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

A Streaming Balancing Graph Partitioning Method Oriented to Memory Computing

A technology of memory computing and balanced graph, applied in computing, resource allocation, program control design, etc., can solve the problems of low quality of division, ignoring characteristics, difficulty in meeting the requirements of division efficiency and division quality at the same time, and achieves obvious effects. Effects of reduced traffic and reduced edge cutting rate

Inactive Publication Date: 2020-01-03
CHONGQING UNIV
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The existing working methods of graph partitioning, whether centralized or stream partitioning, are difficult to meet the requirements of partitioning efficiency and partitioning quality at the same time
For example, the classic Kernighan-Lin algorithm is only suitable for processing small-scale graphs, and the processing efficiency for large graphs is very low, and it is difficult to meet the needs of large graph division and calculation in the era of big data. The information used is usually relatively simple data such as point adjacency, so the efficiency of streaming partitioning is higher than that of centralized partitioning. However, the quality of partitioning is much lower than that of centralized partitioning algorithms.
[0005] In addition, the existing methods ignore the characteristics of the structure of the graph itself, especially most of the real graph structures are characterized by power-law distribution, that is, the nature of "small world".

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
  • A Streaming Balancing Graph Partitioning Method Oriented to Memory Computing
  • A Streaming Balancing Graph Partitioning Method Oriented to Memory Computing
  • A Streaming Balancing Graph Partitioning Method Oriented to Memory Computing

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0043] refer to figure 2 The division results of different division algorithms in Figure Amazon0312, the ordinate indicates the edge cutting rate Ecuti,j Indicates the number of cutting edges between different processing units, and E indicates the total number of edges in the graph. The abscissa indicates different division methods, hash algorithm (H), balance algorithm (B), exponential weight deterministic greedy algorithm (EDG), exponential weight triangular algorithm (ET), linear weight deterministic greedy algorithm (LDG), linear Weighted Triangulation Algorithm (LT), Triangulation Algorithm (T), Blocking Algorithm (C). K=32, the division result 0.08% that red line represents Metis, the numerical value that marks on the figure represents the balance coefficient of corresponding algorithm, and the balance coefficient of triangle algorithm (T) is maximum 2.11, and the balance coefficient of the present invention is 1.00, and the edge cutting rate is 21% , the closest to ...

Embodiment 2

[0045] refer to Figure 3A with Figure 3B , the results of the theoretical power-law graph (PL) and non-power-law graph (ER), the ordinate represents the edge cutting rate The horizontal black line represents the metis division result (ideal value). The closer the value is to Metis, the better the quality of the division. The frequency diagram of the corresponding degree is shown in the upper right corner. The linear weight deterministic greedy algorithm (LDG) division quality in the division result of the ER diagram The best result is 83%, the hash algorithm (hash) result is the worst 92%, the present invention is 85%, and Metis is 81%. In the division results of PL, the edge cutting rate of the present invention is 49%, and that of Metis is 50%. Because the "small world" characteristic of the power-law network graph is particularly suitable for initial community detection, the present invention is particularly suitable for the division of power-law graphs.

Embodiment 3

[0047] refer to Figure 4 In the first step of the present invention, for the label selection problem of points, if there are more than one type of labels in the surrounding neighbors, one is randomly selected at this time, which contains randomness. Result has influence. Come verification stability of the present invention with 3 real figures (amazon0505, amazon0312, amazon0601), ordinate is standard deviation value Diferenc=λ t -λ t+1 Represents the difference between the edge cutting rate of the previous experiment and the edge cutting rate of the next time, and the abscissa indicates the number of experiments, by Figure 4 It can be seen that each time the difference fluctuates at a*10 2 , near (a∈[1,9]), the volatility is very small, which is negligible for large graphs, which shows the stability of this algorithm.

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 stream type equilibrium diagram division method and belongs to the field of parallel diagram calculation. According to an algorithm, for a characteristic of power-law distribution of most diagrams in reality, firstly an original input diagram is quickly subjected to community detection under a constraint condition; secondly communities are sorted; and finally vertexes are allocated to a specified processing unit in sequence according to a deterministic greedy flow algorithm. Compared with a conventional division algorithm, the division quality and the division efficiency are greatly improved. The stream type equilibrium diagram division method is suitable for task division in parallel computing of large-scale networks of power-law distribution.

Description

technical field [0001] The invention belongs to the field of task division of memory computing, and relates to large-scale computing task parallelization and task allocation methods. Background technique [0002] Graph models can express complex relationships between objects and are widely used in many fields, such as social networks, health care, biology, science, and computing tasks can be represented and calculated using graphs. However, as the scale of the problem expands, for example, Facebook has more than 800 million active users and tens of billions of relationship links, so this kind of problem is difficult to calculate on a single computer, and it is necessary to divide the tasks into a parallel computing environment implement. [0003] In response to the analysis requirements of large-scale computing tasks, distributed iterative processing systems have been developed accordingly, such as spark, pregel, GiraphLab, Trinity, etc. These systems mainly use the hash m...

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): G06F9/38G06F9/50
CPCG06F9/3885G06F9/5016
Inventor 钟将李琪
Owner CHONGQING UNIV
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