Supercomputing Goes Data-Intensive

Michael L. Norman

Traditionally, supercomputers have been designed to simulate natural and engineered systems. Grossly speaking such applications turn a small amount of input data into a large amount of output data (e.g., climate simulations). Data-intensive applications are the other way around: one wants to sift through large amounts of data to find a small amount of data (e.g., database query). In Feb. 2012 SDSC put into production the first supercomputer specifically designed for data-intensive applications. Named Gordon (gordon.sdsc.edu), the system incorporates 300 TB of flash solid state disk memory, giving it unprecedented power for large database and data mining applications. It also features 2 TB virtual shared memory "supernodes" to simplify application development. I will describe some of the science applications running on Gordon and highlight research opportunities. UC academic researchers may request time on Gordon via the NSF XSEDE program (www.xsede.org).

ICCS 2012 is organised by