Soft Real - Time Scheduling on a Simultaneous Multithreaded

Simultaneous Multithreading (SMT) is a recently proposed technique that seeks to improve processor throughput by allowing the processor to run multiple threads simultaneously. Soft real-time workloads such as multimedia applications are becoming increasingly important and would benefit from the throughput of SMT processors. However, the scheduling of such workloads on SMT processors introduces new challenges that cannot be met by traditional real-time scheduling algorithms. In particular, to run these workloads on SMT processors we must determine 1) which threads to run simultaneously (the co-schedule) and 2) how to share processor resources amongst them, given the real-time constraints of the applications. We explore several algorithms for co-schedule selection and resource sharing, focusing more on the former. We examine existing multiprocessor scheduling algorithms, including both partitioning algorithms and global scheduling algorithms. We also propose and evaluate some variations on those algorithms that try to utilize the SMT more efficiently. We find that global scheduling algorithms outperform partitioning algorithms due to their ability to migrate tasks from one context to another, which has no associated cost on an SMT. Variations on the algorithms that try to utilize the SMT more efficiently are effective, helping most for task-sets consisting of tasks with similar utilizations. There is a trade off between algorithmic complexity, profiling overhead, ability to provide admission control and schedulability of the algorithms.