The new scheduling challenges posed by chip multicore architectures have opened new research directions in operating system scheduling. This survey paper provides an excellent introduction to this fascinating topic for the reader who has some knowledge of scheduling, but who may lack specialist knowledge of chip multicore architectures.

Chip multiprocessors share resources like caches and memory controllers. Conventional schedulers, which treat each core as an independent processor, give rise to poor performance on multicore processors due to contention between threads for shared resources. Thread-level schedulers mitigate this problem.

Two approaches to thread-level scheduling form the main body of the survey. Contention-aware scheduling is used to map threads to cores, so as to avoid unnecessary competition for shared resources by balancing memory-intensive with compute-intensive threads. Cooperative resource scheduling focuses on threads that share resources within a single multithreaded application. While this approach is hampered by the difficulty of monitoring how threads share data, it is likely to become more important with the emergence of parallel algorithms.

The survey concludes with a review of scheduling in Linux and Solaris, and an informative discussion leading to a summary of the ideal scheduler.