Improved task packing for shared resources in multiprocessor real-time systems scheduled by RUN under SBLP

摘要

Reduction to Uniprocessor (RUN) is an algorithm capable of efficiently and optimally scheduling a set of strictly periodic real-time tasks on a multiprocessor platform when tasks do not share any resources but processors. A basic building block of RUN is the servers, which are entities capable of scheduling a sub-set of tasks via EDF. Recently, support for resource sharing has been incorporated into RUN by the Server Based Locking Protocol (SBLP). SBLP accounts for possible interferences of resource locking times into the RUN servers. The protocol performance, however, depends on how tasks are packed into servers. In this work we describe a new task packing heuristic that outperforms those originally proposed for SBLP. Extensive evaluation experiments are conducted and show that gains of up to 25% can be obtained by the proposed heuristic.