Understanding scheduler performance : a feature-based approach

The thread scheduler of an operating system is a performance-critical service for applications. However, general-purpose operating systems' schedulers do not offer the best performance for all applications, despite an increasing complexity in features and heuristics. Understanding the impact of individual features on the performance of applications is a difficult task because of these features' entanglement. We propose a feature-based model for scheduler and an experimental methodology in order to better understand the ins and outs of process scheduling.