1Uporto/ 2University of Wisconsin – Madison
Significant Financial Capital Expenditures for High Performance Computing (HPC) often require purchase and deploy multiple computational systems (aka a cluster) often incur significate costs and may well cost into $10’s of millions of dollars for HPC systems. Knowing capital cost factors require planning and resource allocations to address user demand in computing, we will evaluate alternatives to the capital expensed multi year amortized homogenous HPC systems, specifically yearly upgrades in a heterogeneous compute clusters. Utilizing computational systems of different vintages and vendors, we attempt to optimize our computational infrastructure for price/performance with a goal to provide large RAM task optimization processing on a budget. Specific evaluations on the feasibility of the secondary computational market for Large RAM and I/O footprints systems using real world examples in the Bioinformatics Big Data analysis computing. We validate our method with two purchases of second hand computers for our cluster with the goal of improving the throughput and cost structure for computational task processing and optimizing for memory/cost at scale.
keywords: HPC, heterogeneous computing, Big Data, cost computing, cloud computing
Poster: Second Life Computing