Assessment of the Realtime Preemption Patches (RT-Preempt) and their impact on the general purpose performance of the system

Title

Assessment of the Realtime Preemption Patches (RT-Preempt) and their impact on the general purpose performance of the system

Authors

Arthur Siro, Carsten Emde, Nicholas Mc Guire

Author Information

Research Fellow of DSLab, Lanzhou University

Abstract

With the maturing of the Realtime Preemption Patches (RT-Preempt) and their stepwise integration into the Mainline Linux kernel since version 2.6.18, we set out to answer the questions:

• How good is RT-Preempt with respect to the worst-case latency?

• How expensive is RT-Preempt with respect to a possible performance degradation of the system?

Taking that a lot of the preemption techniques deployed have their origin in scalability demands and not so much in realtime requirements, the most interesting case to look into is related to uni-processors - on these we would expect the worst-case impact of RT-Preempt. To answer the question, we ran an extensive benchmark series on 2.8-GHz P4 and 1-GHz VIA CIII boards, measuring general OS performance parameters as well as the realtime capabilities. For the latter, a trivial parport toggle program was used.

The results show that high-end CPUs are well supported by RT-preempt in general. Low-end systems typically of interest for automation and control, however, still need some work.

In this paper we will outline the method used for evaluation and present the details of the results.

This work was partly supported by the Open Source Automation Development Lab (OSADL).

Keywords

RT-Preempt, Jitter, Performance