High-Performance Cluster Computing (Ames Laboratory, Scalable Computing Lab [DOE])
David Halstead, email@example.com
The Scalable Computing Laboratory in the DOE Ames Laboratory will be showcasing work on assessing and improving communication of real-world parallel HPC applications, and on large cluster computer systems. In addition to the performance evaluation of multiple high-speed dedicated system area networks, we will be presenting an efficient, threaded, message passing benchmark to optimize the utilization of clustered SMP machines. Included in this work is the option of trading latency performance for bandwidth by using data compression. This has particular relevance to cluster computing, in which compute cycles are cheap, but internode and intersite communications are limited. Finally, research into improving real-world HPC application performance with shared memory emulation APIs and lightweight message passing techniques will be presented together with sophisticated parallel resource management tools.