1 / 8

When Don’t I Use MPI?

When Don’t I Use MPI?. Jonathan Berry Scalable Algorithms Department Sandia National Laboratories June 3, 2008. Informatics Datasets Are Different. Informatics : The analysis of datasets arising from “information” sources such as the WWW (not physical simulation) Motivating Applications:

delila
Download Presentation

When Don’t I Use MPI?

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. When Don’t I Use MPI? Jonathan Berry Scalable Algorithms Department Sandia National Laboratories June 3, 2008

  2. Informatics Datasets Are Different • Informatics: The analysis of datasets arising from “information” sources such as the WWW (not physical simulation) • Motivating Applications: • Homeland security • Computer security (DOE emphasis) • Biological networks, etc. “One of the interesting ramifications of the fact that the PageRank calculation converges rapidly is that the web is an expander-like graph” Page, Brin, Motwani,Winograd 1999 From UCSD ‘08 Broder, et al. ‘00 Primary HPC Implication: Any partitioning is “bad”

  3. Joe: 4 proc. Sally: 4 proc. Sue: 4 proc. memory ref. Shared, Hashed Data Informatics Usage Models Can Be Quite Different

  4. Multithreaded Architectures Can Boost Performance • PageRank performs a sequence of matrix-vector multiplications • “NICE” data are “R-MAT” graphs with maximum degree ~1000 • “NASTY” data are R-MAT graphs with maximum degree ~200k • The MTA-2 runs are nearly data agnostic and have ideal speedup through 20p • The end of MTA-2 scaling indicates that algorithmic work is needed (we’ve seen and overcome behavior like this before) 33M vertices, 268M directed edges PageRank time Number of Processors [K. Devine, S. Plimpton, Berry]

  5. MTA/XMT Programming: Use the Compiler • Here, we sum a quantity over the neighbors of one vertex • The removal of the reduction of “sum” prevents a hot spot • This output is from “canal,” an MTA/XMT compiler analysis tool

  6. MTGL ADAPTER We Are Developing The MultiThreaded Graph Library • Enables multithreaded graph algorithms (XMT, SMP, Niagara) • Builds upon community standard (Boost Graph Library) • Abstracts data structures and other application specifics • Hide some shared memory issues • Preserves good multithreaded performance S-T connectivity scaling (MTA-2) SSSP scaling (MTA-2) Solve time (sec) Solve time (sec) MTGL C MTGL C MTA-2 Processors MTA-2 Processors

  7. Current MTGL Algorithms • Connected components (psearch, visit_edges, visit_adj) • Strongly-connected components (psearch) • Maximal independent set (visit_edges) • Typed subgraph isomorphism (psearch, visit_edges) • S-t connectivity (bfs) • Single-source shortest paths (psearch) • Betweenness centrality (bfs-like) • Community detection (all kernels) • Connection subgraphs (bfs, sparse matrix, mt-quicksort) • Find triangles (psearch) • Find assortativity (psearch) • Find modularity (psearch) • PageRank (matvec) • Network Simplex for MaxFlow Under development: • Motif detection • more Berkeley Open-Source Licence pending

  8. Acknowledgements MultiThreading Background Simon Kahan (formerly Cray) Petr Konecny (Google (formerly Cray)) MultiThreading/Distributed Memory Comparisons Karen Devine (Sandia) Steve Plimpton (Sandia) MTGL Algorithm Design and Development Vitus Leung (Sandia) Kamesh Madduri (Georgia Tech.) William McLendon (Sandia) Cynthia Phillips (Sandia) Generic Programming Background Andrew Lumsdaine (Indiana U.) Doug Gregor (Indiana U.)

More Related