Energy - responsiveness tradeoffs for real-time systems with mixed workload

We explore the performance tradeoffs for real-time systems with dynamic voltage scaling (DVS) capability, when the workload includes aperiodic jobs as well as periodic tasks. As opposed to the assumptions of early works on real-tune DVS or nonpower-aware scheduling of hybrid task sets, the settings require the consideration of two often-conflicting objectives: Improving the responsiveness of aperiodic jobs and reducing the energy consumption. We propose the composite metric, energy * average response time, as a performance measure in energy-aware scheduling of hybrid task sets. Then we develop our framework that integrates dynamic reclaiming algorithm (DRA) and total bandwidth server (TBS) mechanism in variable-speed settings. In addition to the static algorithm, we propose basic reclaiming scheme (BRS) and mutual reclaiming scheme (MRS) that enable the reuse of the system slack arising from early task completions. We also present our bandwidth sharing scheme (BSS) that aggressively exploits the bandwidth reserved for TBS to further slow down the periodic tasks. We provide an experimental evaluation of our algorithms under different workloads and speed settings, and show that BSS can provide significant performance improvements when the actual variability in the workload is high.