HP Clusters in Computational Chemistry
This study reveals excellent scaling, independent of architecture, is found for HP clusters (tested up to 128 processors), effectively reducing the overall computational time. For DFT calculations, clusters of Intel® Itanium® 2 servers, such as HP Cluster Platform 6000, have a distinct performance advantage over similar clusters based on Intel® Xeon® processors, which in turn outperform clusters based on AMD® Opteron™ processors.
Cluster performance, even with the fastest computing architecture, is dependent on sufficient memory and I/O. ADF is a memory-intensive application and, for some job types, stores large amounts of data on disk. Using today’s typical configurations, this case study reveals memory was not a bottleneck to performance. I/O, on the other hand, was improved (from typical Network File System (NFS) configurations) with HP StorageWorks Scalable File Share (SFS)—delivering the best I/O performance, especially useful in large cluster configurations.
Performance on clusters may also be limited by interconnect speed. In this case study, good parallel speedups of clusters with more than four processors came from using high performance interconnects. The parallel efficiency of a particular ADF job depends on many factors. As a guide, this case study characterized the effect of a number of key cluster configuration variables on cluster performance. It also represents part of the collective analyses and understanding that HP and its partners develop in better supporting its computing customers in life and materials sciences.
Get pdf HP Clusters in Computational Chemistry
