Remember back when making your application run faster meant just getting a better computer? While some of us feel nostalgia for a simpler time, in a world of GPUs, fast interconnects, and distributed memory processing, parallelization is pretty much the only game in town for the performance challenged. Relentless gains in processor speed have seen clusters of commodity computers replace the specialty architectures of just a few years ago. Much as traditional operating systems marshal the resources of a single computer, cluster middleware like MPI, workload schedulers and resource managers have in a sense become the new operating system, enabling applications to harness the power of distributed clusters.

While cluster middleware is essential for high-performance environments, it can be complicated. Up until recently, installing and managing a state-of-the-art HPC cluster probably required help from a competent Linux administrator - not a problem if you work in a government lab or multi-national, but definitely an issue if you’re a lone researcher with limited interest in the finer points of cluster administration. While pre-assembled clusters and installation toolkits help with initial installation, they do little to address the bigger challenge of on-going management. Clusters work initially, but they are “brittle." Upgrade an OFED driver, apply a patch, or change the hardware in any way, and all bets are off. Administrators can spend hours if not days tweaking configuration files and navigating a “rats-nest” of software interdependencies. As any IT administrator knows, ongoing management is where the real cost and complexity lies. In fact in their most recent November 2009 study of the HPC market, IDC cited complexity of cluster management and parallel software as the number one “pain points”[1] confronting HPC cluster users.

Clearly the industry needs a better solution, and Dell has one! Click below to listen to a recently recorded Webinar to learn how we have radically simplified both using and managing the industry’s most advanced parallel HPC clusters.

http://www.platform.com/eforums/eforum.asp?1-1MB8DV

About Dr. William Lu
William Lu is leading a team of solution architects helping customers optimize design infrastructures using Platform technology. He and his team have been working with top commercial companies and leading research institutions in implementing grids and cloud solutions. During the past 13 years at Platform, William has worked in product development, professional services, solution architecture, and marketing. Before Platform, William spent 4 years on high-performance computing at CERN and University of Texas after obtaining a Ph.D. degree on high energy physics.