Putting RUN into Practice: Implementation and Evaluation

The Reduction to UNiprocessor (RUN) algorithm represents an original approach to multiprocessor scheduling that exhibits the prerogatives of both global and partitioned algorithms, without incurring the respective drawbacks. As an interesting trait, RUN promises to reduce the amount of migration interference. However, RUN has also raised some concerns on the complexity and specialization of its run-time support. To the best of our knowledge, no practical implementation and empirical evaluation of RUN have been presented yet, which is rather surprising, given its potential. In this paper we present the first solid implementation of RUN and extensively evaluate its performance against P-EDF and G-EDF, with respect to observed utilization cap, kernel overheads and inter-core interference. Our results show that RUN can be efficiently implemented on top of standard operating system primitives incurring modest overhead and interference, also supporting much higher schedulable utilization than its partitioned and global counterparts.