Fair share scheduler for clustered workstations

Abstract

Share, a high level scheduling mechanism, is currently running on a collection of high end Risc System/600 work stations at the IBM TJ. Watson Research Center. This cluster of work stations provides a numerically intensive computing environment for a number of scientists. The scheduling mechanism was designed to manage across the entire cluster and manipulate process priorities in an attempt to deliver a known proportion of the cpu resources to the active users. Share has been used exclusively to allocate cpu resources and will concentrate on only cpu scheduling.