In distributed system, the matrix and graph computation are both need a lot of communication between each nodes. IMO, there is no way to avoid them. Of course, It could be performed by M/R iterations. But it seems very slow and there's an overhead cost. I think that's why we'd like to survey and consider the BSP (bulk synchronous parallel) model.
- We need to explain theoretically about the BSP and How to apply out project. And, regarding matrix and graph, they are closely connected. I expect the synergy between two. However, I think we should clear the relationship between matrix and graph. and our main goal. Any advices are welcome. On Fri, Sep 18, 2009 at 6:30 PM, Edward J. Yoon <[email protected]> wrote: > Firstly, We need to share our plans and consider about overall architecture. > > What's the BSP? What's the relationship between matrix and graph? > What's the plan of matrix and graph packages? What's the our main > goal? > > On Fri, Sep 18, 2009 at 5:52 PM, Apache Wiki <[email protected]> wrote: >> Dear Wiki user, >> >> You have subscribed to a wiki page or wiki category on "Hama Wiki" for >> change notification. >> >> The following page has been changed by HyunsikChoi: >> http://wiki.apache.org/hama/GraphPackage >> >> New page: >> = The Graph Package (Angrapa) = >> The graph package, called Angrapa, is an large-scale graph data management >> framework for analytical processing. It is still an ongoing project. It will >> employ massive parallelism on Hadoop. It aims to achieve the scalability for >> processing tera bytes or peta bytes graph data. Angrapa will be used in a >> variety of scientific and industrial areas, such as data mining, machine >> learning, information retrieval, bioinformatics, and social networks, >> required to process large-scale graph data. >> >> = Description = >> The graph package is new programming framework for graph processing. >> >> = The Main Goal = >> * Easy APIs familar to graph features >> * Store structure suited to graph data when it comes to considering the >> connectivity of graph data >> * Applying data communication method (i.e., BSP) without deterioration of >> graph data locality >> > > > > -- > Best Regards, Edward J. Yoon @ NHN, corp. > [email protected] > http://blog.udanax.org > -- Best Regards, Edward J. Yoon @ NHN, corp. [email protected] http://blog.udanax.org
