Thread Scheduling in Multi-core Operating Systems

Abstract

In this thesis, we address the problem of schedulers for multi-core architectures from several perspectives: design (simplicity and correctness), performance improvement and the development of application-specific schedulers. The contributions presented are summarized as follows:

• Ipanema, a domain-specific language dedicated to thread schedulers for multi-core architectures. We also implement a new abstraction in the Linux kernel that enables the dynamic addition of schedulers written in Ipanema.
• A series of performance and bug tracking tools. Thanks to these tools, we show that the Linux scheduler, CFS, suffers from a problem related to frequency management on modern processors. We propose a solution to this problem in the form of a patch submitted to the community. This patch allows to significantly improve the performance of numerous applications.
• A scheduler model in the form of a “feature tree”. We implement these features independently in order to offer a new fully modular scheduler. This modular scheduler allows us to study exhaustively the different combinations of features, thus paving the way for the development of application-specific schedulers.