Success Story: Intel and Intel® MPI Library

February 29, 2012 in blog-general, for_industry, success_story

When clients have extremely large-scale computations to do, SciNet recommends that they use Intel® MPI Library, a highly tuned, massively scalable communications library that performs well where competitors have trouble even starting.   And when Intel was looking to further improve the performance of their flagship MPI library, they knew they could turn to SciNet and its massive cluster to serve as the ultimate testbed.

SciNet’s GPC cluster, with four thousand nodes connected by ethernet, is one of the largest flat such networks in the world, and provided an extremely valuable testing facility for tuning the MPI library’s performance on extremely large scales. “Over the years we have worked closely with SciNet to develop and tune our Intel MPI Library symbiotically with their high performance computing applications on their compute platform,” said Sanjiv Shah, Director, Technical Computing Software, SSG, at Intel.

“Over the years we have worked closely with SciNet to develop and tune our Intel MPI Library symbiotically with their high performance computing applications on their compute platform.”

Sanjiv Shah, Director, Technical Computing Software, SSG, Intel Corporation.

“ As SciNet’s computational infrastructure and demands from their research community have grown, we have worked closely to ensure that Intel software development tools continue to scale with their computational demands.”

The Intel team used up to 30,000 processors at a time on SciNet’s GPC cluster to test and tune the latest version of their MPI library, released as Intel® MPI Library v4.0, and SciNet users used an early version of the library to run one of Canada’s largest simulations.

But the demand for even more powerful computers and extreme scaling never ends, and so development, and the partnership, continues.   “We’re delighted to continue our fruitful partnership with Intel, helping develop and test their massively scalable MPI library implementation”, said Dr. Daniel Gruner, SciNet’s CTO-Software. “Intel has become a very important software company, taking its cues from the needs of high performance users, such as those using the SciNet systems who are always pushing the envelope for computational science.”