ScalaTrace: Scalable Compression and Timed Replay of Communication Traces

Date:Wednesday, April 15, 2009
Time:11:00am - 12:00pm
Location: LBNL Bldg. 50F, Room 1647

Speaker:
Frank Mueller
Department of Computer Science
North Carolina State University
   

Abstract:

    Characterizing the communication behavior of large-scale
    applications is a difficult and costly task due to code/system
    complexity and their long execution times. An alternative to running
    actual codes is to gather their communication traces and then replay
    them, which facilitates application tuning and future procurements.
    While past approaches lacked lossless scalable trace collection, we
    contribute an approach that provides orders of magnitude smaller, if
    not near constant-size, communication traces regardless of the
    number of nodes while preserving structural information. We
    introduce intra- and inter-node compression techniques of MPI
    events, we develop a scheme to preserve time and causality of
    communication events, and we present results of our implementation
    for BlueGene/L. Given this novel capability, we discuss its impact
    on communication tuning and beyond. To the best of our knowledge,
    such a concise representation of MPI traces in a scalable manner
    combined with time-preserving deterministic MPI call replay are
    without any precedence.

Host of Seminar:

    Paul Hargrove