Cluster Computing - Preliminary List of Projects

Search the literature on one of the following topics and produce a 10 to 15 page report with a summary of the reference material and URL or citation of the literature. Remember to correctly credit the source for any material.

Also produce Powerpoint slides for a 30 to 40 minute talk on the subject. Again credit any sources.

As sources for references use scholar.google.com and library databases in addition to normal web searches.

Research Projects

For each of these topics, you should discuss the raw performance attainable, that over TCP/IP, that over any native API, and that of MPI over any of these available. You should also summarize any research which has been done to optimize the performance of MPI collective operations and MPI-2 one sided communications (RMA). You should also discuss the implementation details where these are relevant to the performance.
  1. Checkpoint and Restart facilities and mechanisms in LAM
  2. Optimization of Collective Communications in MPI
  3. MPI Performance Performance and Optimizations on Myrinet
  4. MPI Performance Performance and Optimizations on Infiniband
  5. MPI Performance Performance and Optimizations on Gigabit/10 Gigabit Ethernet
  6. Cluster Performance Monitoring (NWPerf, PAPI)
  7. System overhead factors effecting performance on Very Large clusters
  8. Single System Image for Cluster Computing (OpenSSI)
  9. High Availibility Linux Systems and Clusters (High availability Oscar)
  10. Parallel I/O and the Parallel Virtual File System (PVFS)

Implementation and Research Projects

  1. Deploy the Automatically Tuned Linear Algebra Software (ATLAS) software package and document the performance improvements achieved using it.
  2. Deploy ScaLAPACK and test it and write a report on the process and performance.

farrell@cs.kent.edu