Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
CGraph: a distributed storage and processing system for concurrent iterative graph analysis jobs
Zhang Y., Zhao J., Liao X., Jin H., Gu L., Liu H., He B., He L. ACM Transactions on Storage15 (2):1-26,2019.Type:Article
Date Reviewed: Oct 4 2019

There are many distributed (and commercial) graph processing platforms (for example, services related to social networks) that “need to handle massive concurrent iterative graph processing (CGP) jobs”--processing on large-scale graphs. Recent methods for supporting graph analytics on distributed environments (especially if high performance is expected) include data locality, accelerated state propagation, balanced load, and reduced consumption of memory. However, if many CGP tasks are running on one graph and each CGP job accesses the processed graph, the same data is loaded many times into the (cache) memory. As a result, low throughput is observed. Thus, how to improve the throughput of CGP jobs is of fundamental importance.

A simple observation can be made: CGP jobs traverse each vertex for their own purpose in processed graphs, but there are many intersections among these vertexes that can be described by the spatial correlation of data accesses. The same can be observed with regard to time domain--as a result, we have a temporal correlation of data accesses. The authors propose a processing system (CGraph) that supports CGP jobs and increases the expected efficiency. It is based on an approach where the graph structure is decoupled into a data-centric model; the shared parts of the graph structure are streamed into the cache, triggering CGP jobs. They are also allowed to share copies of the graph structure data in the memory and their access. A new communication scheme with reduced communication costs is also proposed, based on batches related to the distribution characteristic of communication. Moreover, a novel “efficient incremental load-balancing strategy is employed for the CGP jobs.”

Finally, the paper compares CGraph performance to other distributed graph processing systems (Gemini and Seraph); experiments show that “CGraph performs much better.” The authors propose further work to improve the results of throughput via new storage technologies (GraFBoost) and new processing technologies based on graphics processing unit (GPU)-based and ASIC accelerators.

Reviewer:  Dominik Strzalka Review #: CR146715 (1912-0435)
Bookmark and Share
  Featured Reviewer  
 
General (C.0 )
 
 
Architectures (H.5.4 ... )
 
Would you recommend this review?
yes
no
Other reviews under "General": Date
Structured computer organization (3rd ed.)
Tanenbaum A., Prentice-Hall, Inc., Upper Saddle River, NJ, 1989. Type: Book (9780138546625)
Oct 1 1991
Principles of computer systems
Karam G., Bryant J., Prentice-Hall, Inc., Upper Saddle River, NJ, 1992. Type: Book (9780131594685)
Sep 1 1992
Computer organization
Scragg G., McGraw-Hill, Inc., New York, NY, 1992. Type: Book (9780070558434)
May 1 1994
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy