CLOUD

Under the Debugger Software Enhancement program for Petascale production grade tools, awarded to Allinea Software Inc. by Oak Ridge National Laboratory in Q2 2009, Allinea’s Distributed Debugging Tool (DDT) is setting new levels of debugger scalability on Jaguar, a Cray XT5 and one of the world's largest supercomputers.

 

In Q2 2009, Allinea began a collaborative project with ORNL to extent the scalability of its DDT product. The project goal is to enable ORNL's users to debug MPI applications that span many hundreds of thousands of processors, while delivering novel capabilities that can radically simplify this task. Following successful completion of the early stages of the project, Allinea was able to demonstrate that DDT can debug a 220,000 process application running on ORNL's Jaguar supercomputer.

 

"Given the inherent complexities of developing applications at Petascale, it is very important that our users are not frustrated by the very tools that are intended to help solve their problems," said Dr David Lecomber, CTO of Allinea. "Our initial work has therefore focused on making the basic Petascale debugging experience much the same as it would be on very modest numbers of processes. I am pleased to say that we can now launch a debugging session at 220,000 processors in little more time than it takes to spawn the application itself. Current benchmarks also show that we can perform key actions – like stepping 220,000 MPI processes, setting breakpoints, or comparing variables across this number of processors – in a couple of hundred milliseconds or less. This is a massive step beyond what was previously possible in MPI debugging. We are delighted to have been able to achieve this result in such a short period of time."

 

The critical factors that have permitted Allinea DDT to scale past original project goals are attributed to an excellent collaboration with ORNL and the underlying design of the product, which lends itself well to incremental, modular improvements.

 

"ORNL and Allinea are partnering to enhance the scalability of the DDT debugger with the goal to support the complete Jaguar System. The work has progressed in a timely manner and has demonstrated the ability to debug a 220,000 process job,” commented Richard L. Graham, Applications Performance Tools Group Leader at ORNL. “We are very pleased with our partnership and the success we are achieving with Allinea Software. Our collaboration with Allinea Software is delivering excellent results."

 

Allinea Software will be exhibiting at the Supercomputing Conference (SC’09) in Portland, USA from November 16 -20, 2009, booth #1808.