Projects

• choose a project
• prepare a problem description and an approach for solving the problem
• prepare progress reports
• prepare a final report, demonstrate and deliver software

Each project includes a) benchmarking of some existing code and b) a technical problem and/or tool evaluation with the potential to extend existing work.

Benchmarks:

• choose one that has support for both MPI and OpenMP (no stress tests)
• ASCI Purple Benchmarks for MPI and MPI+OpenMP
• More benchmarks for reference (you may NOT choose one of these):
• NAS Parallel Benchmarks for SMPs (OpenMP)

Project Topics:

1. Anwar Ali, Nhon Van Nguyen, Annika Edwards
   MPIP adaptation for i32/Linux, evaluation and enhancements.
   Benchmark: IRS
   Project Web: http://www4.ncsu.edu:8030/~aredward/csc591c/
   
   
2. Anubhav Dhoot, Kunal Shah
   DPCL installation for i32/Linux, evaluation and demonstration on ASCI benchmarks.
   Benchmark: smg2000
   Project Web: http://www4.ncsu.edu:8030/~kdshah/report.htm
   
   
3. Vikram S. Poojary, Raj Kumar Nagarajan
   Paradyn installation for i32/Linux, evaluation and demonstration on ASCI benchmarks.
   Benchmark: AZTEC
   Project Web: http://www4.ncsu.edu:8030/~rknagara/work/cluster_project.htm
   
   
4. Harini Ramaprasad, Salil Pant, Sibin Mohan
   Tau installation for i32/Linux, evaluation and demonstration on ASCI benchmarks.
   Benchmark: ParBenCCh
   Project Web: http://www4.ncsu.edu:8030/~smohan/ClusterWebPage.htm
   
   
5. Frank Castaneda, Nikola Vouk
   PAPI evaluation for threading and HW counter multiplexing: limits in MPI and OpenMP frameworks as well as mixes, design (and, if time, implementation) of methods to extend the functionality to circumvent problems.
   Benchmark: Info Needed
   Project Web: http://www4.ncsu.edu/~nvouk/exploitinghyper.html
   
   
6. Jadeep Marathe
   Address Trace Generation for OpenMP Programs using Dynamic Instrumentation
   Project Web: http://www4.ncsu.edu/~jpmarath/index.html
   
   
7. FT-MPI installation for i32/Linux, evaluation and demonstration on ASCI benchmarks.

More details:

1. ASCI benchmarks: This is independent of the projects. The objective is to gain experience with larger MPI/OpenMP applications.
2. You have to install the tool and demonstrate its use for your chosen ASCI benchmark (among other, smaller tests). This requires that you
   1. write some test programs yourself to understand how the tools work
   2. integrate similar code into the ASCI benchmark to trigger the functionality of the tool
   3. evaluate and interpret the result
   4. find ways to improve the performance of the application and try them out (again evaluating results with the tool)
   5. suggest ways of improving the tool (wish list concentrating not on GUIs but on new functionality or runtime support)
   6. if there's time, prototype an improvement (e.g., this should definitely be done for MPIP)

Other Unassigned Ideas:

• hyperthreading+linux scheduler+mpi (need Intel Pentium4 or Xeon with Hyperthreading for this; if some student has one at home, the problem can be assigned; requires Linux kernel modifications)
• ATLAS assessment and improvement for AMD Athlons

Other Pointers:

You may find that benchmark codes require any of the libraries below. Notice that the first two are already installed (under /opt) while the last would have to be compiled by you. Other libraries can be added upon request if you provide a compiled version ready to install.

• BLAS
• LAPACK
• SCALAPACK