Operating System Suites for Petascale Systems

Pete Beckman

In days of old, supercomputers ran a simple operating system. As MPPs and clusters emerged, they began to run thousands of instances of an operating system. The largest scale supercomputers now run four or more operating systems on different functional units within the machine, and this trend is likely to continue. These collections of operating systems, each designed for specialized capabilities, form the "OS Suite" for supercomputers. Most of the time, they are based on Linux. However, as machines scale up, simple design choices what worked well for servers or a thousand nodes, suddenly begin to become bottlenecks. This talk will focus on the new areas of research for OS Suites such as those running on Cray's Red Storm and IBM's BG/L.