|Name:||Future Trends(5) Mapping Biomedical HPC Workloads to Low Power SoC Environments|
|Time:||Monday, June 17, 2013
1:45 PM - 1:50 PM
|Room:||Multi-Purpose Area 4 (MPA 4)
CCL - Congress Center Leipzig
|Speakers:||Shun Liang, University of Cambridge|
|Abstract:||The EC EuroCloud project proposes a “Server-on-Chip”(SoC) to provide a very dense low power server, for data centres and data intensive research environments.
At the same time leading edge cancer research is increasingly underpinned by “Big Data” analysis workloads driven by the explosion of genomic and imaging petascale datasets routinely available to researchers. The Cancer Research UK Cambridge Institute has joined the Eurocloud project to see if there is a potential benefit of the approach in this area, using novel low-energy cloud centred hardware, based on the Calxeda EnergyCore 32 bit ARM-based SoC.
We expected that, like other typical Cloud environments, this architecture would have a problem with “Big Data”. But a key feature of the EnergyCore is integrated 80Gb high performance networking; additionally, in our current system each 2u component provides 6TB of storage, through 24 SATA connections. This environment can support a parallel file system, in a module which can also provide 192 low energy compute cores. Our proposed application architecture, with LAMP services, parallel scheduler and parallel file system in a single modular component may provide an ideal solution for many traditional data-centric HPC workloads.
This solution, or its 64 bit successors, may profoundly impact the model for HPC provision in an area increasingly important to medical research and clinical practice.
Mark Calleja, University of Cambridge; Shun Liang, University of Cambridge; Peter Maccallum, University of Cambridge; Anne Pajon, University of Cambridge