May 2007

Parallel IO Optimization for Lustre File System

This week, at the Seventh IEEE International Symposium on Cluster Computing and the Grid, Brazil, the paper \Exploiting Lustre File Joining for Effective Collective IO\ by Weikuan Yu, Jeffrey S Vetter, Shane Canon and Song Jiang was presented. The paper exploits the use of Lustre File Joining to create MPI-IO files with adaptive striping width, in order to minimize the use of Lustre OSTs while achieving comparable aggregated IO bandwidth. Further studies on its applicability on Cray XT and its benefits to cluster-wise energy consumption are under way.


Intel Woodcrest Evaluation for Scientific Computing

A paper titled \Intel Woodcrest: An Evaluation for Scientific Computing\ by Philip C. Roth and Jeffrey S. Vetter was presented at the 8th LCI International Conference on High-Performance Clustered Computing. The paper explores the performance of the Intel 5100-series dual-core processor (code-named \Woodcrest\) using a workload of computational and communication microbenchmarks, kernels, and full scientific applications from areas like climate and molecular biology.


FT FPGA research achieves 100X speedup for Human Genome Sequencing on Cray XD1

A paper entitled \Performance Evaluation of Biological Applications that use FPGAs\ by Olaf Storaasli and Weikuan Yu (ORNL), Dave Strenski (Cray) and Jim Maltby (Mitrionics) presented May 7th at the Cray Users Group, documents how 100X speedups (over a 2.2 GHz Opteron) were achieved on a Cray XD1 Virtex 4 FPGA for comprehensive DNA, RNA and protein sequencing applications. Key to this success was that 98.6% of the algorithm's computations are concentrated in a kernel that was computed in parallel during each FPGA clock cycle.