Title: Department of Computer Science Seminar Date: Monday, 5 May 2003 Time: 3:00 pm to 4:00 pm Venue: Room N101, CSIT Building [108] Title: Enabling the Efficient Use of SMP Clusters: The GAMESS/DDI Model Speaker: Ryan Olson (Dept of Chemistry, Iowa State University) Abstract: GAMESS is a widely used computational chemistry package. Depending on the problem at hand it is not uncommon for GAMESS calculations to run for days, weeks or even months on state of the art single processor workstations. To reduce the time to solution to more acceptably levels and/or to enable even larger problems to be tackled, over the years much effort has been directed at enabling GAMESS to run on various high performance parallel computing platforms. This effort has seen the code move from the use of early message passing models, to MPI, to the Distributed Data Interface (DDI). In recent years, as technology has trickled down from supercomputers into the personal computer market, clusters of PC's (Beowulf clusters) have become a popular and relatively inexpensive means of accessing serious computational power and very large aggregate memory. Originally these clusters comprised single processor boxes coupled via ethernet, but are now increasingly assembled from small shared memory parallel (SMP) nodes containing 2, 4, 8 or more CPUs linked via a choice of various advanced network technologies (possibly with "intelligent" interface cards). The aim of this seminar will be to outline our efforts to exploit these new environments. The talk will give an outline of the DDI and the DDI programming model. We will discuss the initial implementation of DDI that was on the Cray T3D system using Cray's SHMEM library, then outline how the model was extended to support Beowulf Cluster type systems. We will then consider our recent work to extend the DDI model for clusters of SMP systems and large NUMA based SMPs. Performance data will be presented for a variety of systems, including both the HP SC and GS systems available at the ANU in the APAC National Facility. Biography: Ryan Olson, Dept of Chemistry, Iowa State University and The Scalable Computing Laboratory, Ames, IA 50011 URL: http://cs.anu.edu.au/lib/seminars/seminars03/dept20030505