QoS aware dynamic time-slice tuning

The ability to manage quality of service (QoS) and to provide service differentiation has been very important in a wide range of computing environments [4], [3], [5]. In modern operating systems, multiple applications share processor cores and take turns to execute. Each application typically runs for a while before its CPU time-slice (allocated quantum) expires or the execution is blocked due to I/O operations etc. This gives users the transparent feeling of simultaneously executions even on single-core machines and is usually referred as time sharing. The actual CPU time-slice for a process is based on a base time-slice and some dynamic adjustments. The base time-slice is the same for processes within the same priority level. Minor runtime adjustment to the base time-slice is possible as per the dynamic characteristics such as average sleep times. On the application side, the process priority and associated time-slice assigned by the operating system (OS) scheduler can usually affect its performance. The performance metric we focus in this work is the IPC (instruction per cycle). When multiple applications compete for the CPU cycles in a round-robin manner, longer CPU time-slice often results in larger fraction of CPU execution and leads to performance speedup for an individual application. CPU-intensive applications are more likely to benefit from a large time-slice as opposed to I/O-intensive applications.