Multicasts for Faster Science Applications on Beowulf Clusters

Other

Scientific paper

Rate now

  [ 0.00 ] – not rated yet Voters 0   Comments 0

Details

Scientific paper

We describe the use of reliable multicasts in existing message passing (MPI) programs to reduce collective communication bottlenecks. By broadcasting information through one point-to-multipoint message instead of a series of (tree-structured) point-to-point TCP messages, communication delays can be reduced by factors up to log2(Nnodes), e.g., 6x for 64-node clusters. This reduction in the cost of collective communications broadens the class of problems appropriate for Beowulf-style supercomputing. We present raw communication and science application (Swift N-body integrator) speed-up results. In addition to impressive gains in scaling for some classes of existing programs, the broader benefit of these results is to reduce the importance of domain decomposition. The cost of sending a message to all nodes is the same as one node-to-node message. This allows Beowulfs to be applied effectively even to problems requiring every node to know the current state of all other nodes' particles/cells/.... This should significantly increase the range of problems approachable with relatively cheap, Ethernet-based Beowulf clusters. The reduced cost of collective communications can also make it easier to write adequate parallel programs, thereby increasing the pool of potential Beowulf programmers. Finally, we discuss as a future possibility how Internet2's ubiquitous support for multicasts and high reliability make it a promising medium for expanding these results to geographically distributed computing resources. This research is supported by NASA AISRP through grant NAG5-9320.

No associations

LandOfFree

Say what you really think

Search LandOfFree.com for scientists and scientific papers. Rate them and share your experience with other people.

Rating

Multicasts for Faster Science Applications on Beowulf Clusters does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.

If you have personal experience with Multicasts for Faster Science Applications on Beowulf Clusters, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Multicasts for Faster Science Applications on Beowulf Clusters will most certainly appreciate the feedback.

Rate now

     

Profile ID: LFWR-SCP-O-1891415

  Search
All data on this website is collected from public sources. Our data reflects the most accurate information available at the time of publication.