ABSTRACT: By scheduling the non real-time tasks earlier while still meeting deadlines for the real-time tasks, the overall system performance can be improved. In particular, we believe that the variability in the execution time requirements of realtime tasks can be effectively leveraged to reduce response times of non real-time tasks. We propose a novel processor sharing algorithm where the processor share of RT jobs increases with their progress based on the empirical probability distribution of execution times of real-time tasks, to adaptively schedule variable requirement real-time to maximize the minimum expected service rate to non real-time tasks at any instant.
(A copy of the slides for the talk presented at the conference is also available.)