The Polling Effect on the Schedulability of Distributed Real-Time Systems

The usage of polling tasks continues to be quite common in today's distributed real-time systems, despite the availability of event-driven software mechanisms and response time analysis techniques that can be applied to this kind of systems. This paper proposes a model for polling tasks that allows current response time analysis techniques for event-driven distributed systems to be applied, and it also studies the impact that polling has in the schedulability of a distributed system, using analytic results. A performance evaluation on an Ada-based platform is also provided. As expected, polling produces response times much higher than a pure event-driven alternative. The analysis techniques and the evaluation presented in the paper allows engineers to assess the negative effect of polling on the schedulability of distributed real-time systems.