High Performance Computing (HPC) and Big Data analysis have several core requirements in common, writes Wikibon CTO and Cofounder David Floyer and Principal Research Contributor Stuart Miniman in “High performance computing meets big data“. These include high processor density, high-speed IO, power density sufficient to support the demanding HTC components, and automated high-availability solutions that can keep production going when a node fails. In both cases the key metric is time-to-value, and the key technical problem is very similar — keeping processors fed with data and avoiding stoppages.

As Big Data analysis moves from development/test to large-scale production environments, it will require all of these of its infrastructure to meet its deadlines and SLAs. Good job schedulers and data consolidation are more important than hypervisors, as the large amounts of data involved and difficulty in moving that data dictates an un-virtualized processing environment. The infrastructure tends to be organized around specific “pools” of data, often in Hadoop clusters.

The Professional Alert looks at systems from four leading vendors: early market leader Dell (R720XD), Hewlett-Packard (Proliant SL4540), Oracle (X3-2), and HPC vendor SGI (InfiniteData Cluster). The Dell and HP solutions are not fully integrated as delivered. The analysis rated the SGI InfiniteData Cluster highest in scalability and core and storage density and lowest in three-year environmental costs in part because of its higher density and high level of integration.

The authors recommend that companies building in-house large-scale production Hadoop infrastructures choose pre-integrated solutions with optimized data center resources and suggest that SGI be included in all Hadoop production architecture evaluations.

As with all Wikibon research, this full Professional Alert is available without charge on the Wikibon Web site. IT professionals are invited to register to join the Wikibon community, which allows them to post comments on published research, questions, tips, Professional Alerts and longer papers.