August 15, 2006
[ return to the news list ]
HPC Wire (05/19/06) Vol. 15, No. 20,Hoeflinger, Jay; Meadows, Larry
OpenMP programs cost less to program and maintain, but OpenMP's reliance on a shared memory has kept its use restricted to a single multiprocessor machine. Cluster OpenMP from Intel addresses this limitation by facilitating the operation of an OpenMP program across a cluster of multiprocessors, using a software layer that deploys a distributed shared memory (DSM). Cluster OpenMP thus offers a cheaper hardware purchase and a cheaper option for programming. Cluster OpenMP uses as its foundation an exclusively licensed version of the TreadMarks DSM system, augmented to accommodate bigger volumes of sharable data, a higher number of processors, multiple threads per process, and the ability to function on modern cluster interconnects. A wide swath of applications are naturally well-suited to Cluster OpenMP, such as those that execute rendering, data-mining, parallel search, speech and visual recognition, and genetic sequencing. Any application that utilizes a large volume of read-only sharable data and only a small amount of read/write sharable data while employing synchronization lightly is a potentially good Cluster OpenMP candidate. The identification of sharable data is crucial to adjusting the performance of OpenMP programs, since it reveals insights into data access properties that are often concealed because of OpenMP's procedural focus.
For the complete article, see http://www.hpcwire.com/hpc/658711.html.
Posted by rab5 at 09:14 PM
[ return to the news list ]
