Compensating for Timing Jitter in Computing Systems with General-Purpose Operating Systems

Fault-tolerant frameworks for large scale computing clusters require sensor programs, which are executed periodically to facilitate performance and fault management. By construction, these clusters use general purpose operating systems such as Linux that are built for best average case performance and do not provide deterministic scheduling guarantees. Consequently, periodic applications show jitter in execution times relative to the expected execution time. Obtaining a deterministic schedule for periodic tasks in general purpose operating systems is difficult without using kernel-level modifications such as RTAI and RTLinux. However, due to performance and administrative issues kernel modification cannot be used in all scenarios. In this paper, we address the problem of jitter compensation for periodic tasks that cannot rely on modifying the operating system kernel. ; Towards that, (a) we present motivating examples; (b) we present a feedback controller based approach that runs in the user space and actively compensates periodic schedule based on past jitter; This approach is platform-agnostic i.e. it can be used in different operating systems without modification; and (c) we show through analysis and experiments that this approach is platform-agnostic i.e. it can be used in different operating systems without modification and also that it maintains a stable system with bounded total jitter.