ECONOMICS

What happens when Alliance Grid Testbed institutions join forces for a workshop over the AG and the AGT? Answer: They pull off a successful, cutting-edge MPI (Globus Version) over the Alliance Grid Testbed workshop. Hosted by OSC (Ohio Supercomputer Center) and sponsored by the Alliance PACS (Partners for Advanced Computational Services), the workshop was groundbreaking for testing and running code on the Alliance Grid Testbed (AGT). The AGT consists of nationally distributed clusters with 16 – 32 compute nodes, dual 2.4 GHz Pentium 4 Xeon processors, 2 GB of RAM, 60 GB of local scratch space, Gigabit Ethernet, one head node, and a storage node. “The work that has been done by the AGT sites over the last six months has been worth noting– but the work that they’ve done during the last few weeks has been impressive. We have gone into uncharted territory and have made things work well.” said Leslie Southern, the PACS Training Group Lead. The core components of the workshop lecture were overviews of the AGT, the Globus Toolkit, and MPICH-G2. Some of the workshop topics included: •The Alliance Grid Test bed (AGT) • What it supports • What it explores • Hardware •The Globus ToolKit (GTK) •Security •Resource Management •Data Transfer •mpirun command •Format of machine file •MPI - Globus Version (MPICH-G2) • Resource Specification • RSL • How to Run a Program • Improving Remote Data Transfers • MPI Communication Strategy • Grid Topology • Client / Server Functions Quite a bit of time and effort went into making the workshop a success. “The person most instrumental in the success of this workshop was Jim Giuliani. He leads the design and administration of our AGT cluster at OSC. He was instrumental in pulling together the technical aspects of the workshop, from setting up the 90 workshop accounts to testing among the AGT sites,” said Dr. David Ennis, workshop instructor and Senior Supercomputing Resource Specialist. “We were also fortunate to benefit from the expertise and efforts of the cluster administrators at the other AGT sites. The workshop was an example of collaboration at its best,” said Ennis. The workshop attendees valued not only the lecture, but also the hands-on portion of the workshop. “We really covered the nuts and bolts of submitting jobs and using distributed resources. The venue (Access Grid) was the next best thing to being there… We could hear each other’s questions and answer them from our own experiences! We gained experience that would otherwise have been impossible,” said Sue Brown, from the University of Hawaii. The AGT institutions include Argonne National Laboratory, Boston University, Caltech, HPC-UNM, University of Kentucky, NCSA, OSC, University of Wisconsin. Other sites participating in the workshop included North Dakota State University, Dartmouth College, Purdue University, University at Buffalo, University of Hawaii, Los Alamos National Laboratory, San Diego Supercomputing Center, and Pittsburgh Supercomputing Center. Almost 100 participants attended from the various sites.