CAPACITY PLANNING FOR SOLARIS~R SRM: 'ALL I EVER WANT IS MY UNFAIR ADVANTAGE' (AND WHY YOU CAN'T HAVE IT)

摘要

Traditional UNIX and MVS time-share schedulers attempt to be fair to users by virtue of using a round-robin style algorithm for allocating CPU and other system resources. However, there is an intrinsic loophole whereby the scheduler can be forced to favor a greedy user who runs many short CPU-time processes. This loophole is not a pathology but a necessary consequence of ensuring system responsiveness to small interactive CPU demands. The Solaris? System Resource Manager (SRM) introduces a new scheduling paradigm that constrains the scheduler to be "fair" to all users; independent of the number of small processes owned by any one user. The SRM paradigm draws on the concept of allocating resource "shares" to groups and users. In this paper we offer capacity planning guidelines for SRM based on several share allocation scenarios for groups of users. A very brief comparison with IBM's workload manager (WLM) is also given at the end of this paper.