The Effects of Precedence and Priority Constraints on the Performance of Scan Scheduling for Hypercube Multiprocessors

In the absence of scheduling constraints, Scan scheduling has been shown to considerably improve performance in hypercube multiprocessors relative to previously studied job scheduling disciplines. In practice, jobs may have properties that constrain the choices a job scheduler can make, potentially limiting the scheduler's ability to affect performance. Two common types of job scheduling constraints are precedence and priority constraints. We examine the effects of these constraints on the ability of Scan to improve performance in hypercube systems. We find that, even under severe constraints, Scan scheduling retains its large performance advantage.