Operating system boosts high performance computing
2009 R&D 100 Winner
The increase in computation power and density from multi-core processors is encouraging for researchers and designers, but the majority of scientific parallel computing applications depend as much on memory subsystem performance as on compute performance. In addition, the message-passing-based model upon which nearly all scalable parallel applications are based decreases the limited memory bandwidth available to a processor.
In response, some high-performance computing organizations are redesigning and rewriting existing parallel applications using alternatives to the existing programming model. The Catamount N-Way (CNW) Lightweight Kernel, operating system from Sandia National Laboratories, Albuquerque, N.M. and Operating Systems Research, Washington, D.C., provides an alternate solution.
Using the existing features of a multi-core processor to cut the memory bandwidth requirement of message-passing-based applications in half, CNW provides new capabilities that significantly increase the raw performance of critical message-passing operations. CNW cuts the required memory bandwidth for intra-node message passing in half. It enables in-place collective operations, threaded reduction operations, and high-performance, one-sided data movement operations in approximately 25 lines of operating system code.
Sandia National Laboratories
Operating Systems Research