Nicer CPU Scheduling and Bandwidth Management for Volunteer Computing

Abstract

We proposed methods dealing with performance reduction problem when running volunteer computing application. First, we deal with CPU time allocation problem on CFS-based system. We use adaptive reservation dealing with maintaining CPU time for foreground process. We solved problem of user boundary using system's global statistics. Our method can maintain foreground application's performance and can reclaim more CPU time compared to that of static allocation method. In the second part, we study new HTTP/2. The previous HTTP/1.1 solves problem of unutilized bandwidth by opening multiple connection. However, this can cause network bandwidth contention when bandwidth is limited. In this work, we focus our study on the uses of multiplexed streams. Our result shows that the multiplexed streams can replace multiple seperate connections and is more network friendly to other applications.