OpenMP Programming Contest

SC'05

Sponsored by the OpenMP ARB
 

 

NEW! Here is a serial implementation of the SSCA#2 benchmark derived from John Feo's MTA port. This may assist you in your efforts.
 

The object of the contest is to encourage the use of OpenMP for scalable high-performance computing. The basis of the contest is the DARPA HPCS Scalable Synthetic Application #2, Graph Theory (SSCA2). You will find the following documents useful:

- The specification for the SSCA#2 benchmark, including the written specification and an executable specification written in Matlab (from http://www.highproductivity.org/SSCABmks.htm)

- A parallel implementation of SSCA2 written using pthreads and the SIMPLE library developed by David Bader and Kamesh Madduri (currently of Georgia Tech)

- A paper by Bader and Madduri describing the design and implementation of their solution to the benchmark

 

The object of the contest is to implement this benchmark, following the rules provided in the written specification, in OpenMP. Three prizes will be awarded:

 

First prize: $1000 plus a 60GB iPod.

Second prize: $500 plus a 4GB iPod nano.

Third prize: $250 plus a 1GB iPod shuffle.
 

The entries will be judged by selected members of the OpenMP ARB. Entries must be received no later than midnight Pacific Standard Time, November 5, 2005. Entries should be submitted by e-mail to omp-arb@openmp.org .

 

Prizes will be announced at the OpenMP BOF at SCÂ’05 in Seattle . Entrants need not be present to win. Entrants may enter as a team, but will designate one team member to receive the prize.

 

The criteria for judging are:

- Performance and Scalability: 50%

-  Usage of OpenMP: 50%

 

Performance will be measured on the average of the normalized timings for all four timed kernels. Scalability will be measured on the speedup over a single processor run.

 

Usage of OpenMP is a subjective decision by the judges, but will include factors such as elegance of code and novel use of OpenMP.
 
Performance figures deleted due to my ineptness. Please contact lawrence.f.meadows at Intel dot Com for any questions.