Community Grids Laboratory

1
Community Grids Laboratory

• http//grids.ucs.indiana.edu/ptliupages/presentati
  ons/PC2007/ contains a set of 4 parallel
  computing lectures given by Fox at Microsoft
  Research February 26 to March 1 2007
• http//www.connotea.org/user/crmc is a tagged
  collection of multicore links
• We analyzed Microsoft CCR which supports an
  exchange of messages between threads using named
  ports and a service architecture DSS built on CCR
  for Robotics applications

• DSS gives on multicore systems 40 microsecond
  latency for 2-way messages and CCR a few
  microsecond latency for MPI style communication
• This package is attractive for a broad class of
  applications
• Looking at machine learning and computer chess
  applications

2
Summary of CCR Stage Overheads for Intel 4-core
2-processor Machine

• These are stage switching overheads in
  microseconds for a set of runs with different
  levels of parallelism and different message
  patterns each stage takes about 30 microseconds.
  2-core 2-processor Xeon overheads in parentheses.
  We also benchmarked 2-core 2-processor AMD
  machine
• These measurements are equivalent to MPI
  latencies
• Match uses a number of threads number of
  parallel computations
• Default uses 8 threads