Improving Kernel Scalability by Lock-Aware Thread Migration

Yan Cui and Yu Chen
Advisor: Yuanchun Shi

Tsinghua University, China

Multi-core architecture has been adopted by chip manufacturers. According to Moore's Law, cores in a single chip are still increasing. Within ten years, each chip will accommodate thousands of cores. Operating systems, however, cannot follow the pace of hardware and scale well with the number of cores due to the synchronization primitives protecting the shared data in the kernel. The problem will become more serious in the future many-core platforms.

To address this problem, we propose to improve the parallel scalability of operating system by lock-aware thread migration. This method has several advantages such as shorten lock waiting time, enhanced locality of shared data and evaluation on the real multi-core systems. However, several challenges exist for this method to achieve the potential. To address these problems, we propose several novel methods such as assigning cores executing lock code dynamically, migrating threads according to the hotness of each lock. Preliminary experimental results show that lock-aware thread migration can effectively improve the kernel scalability for a wide range of workloads.

Long Abstract

Back to Program