Energy-Aware Scheduling for Streaming Applications on Chip Multiprocessors

Streaming applications have become increasingly important and widespread, and they will be running on soon- to-be-prevalent chip multiprocessors (CMPs). We address the problem of energy-aware scheduling of streaming applications, which are represented by task graphs, on a CMP using on/off and dynamic voltage scaling (DVS) on a per-processor basis. The goal is to minimize the energy consumption of streaming applications while satisfying two typical quality-of-service (QoS) requirements, namely, throughput and response time. To the best of our knowledge, this paper is the first work to tackle this problem. We make a key observation: the trade-off between static power and dynamic power should play a critical role in both parallel processing and pipelining that are used to reduce energy consumption in the scheduling process. Based on this observation, we propose two scheduling algorithms, Scheduling 1D and Scheduling 2D, for linear and general task graphs, respectively. The proposed algorithms exploit the difference between the two QoS requirements and perform processor allocation, task mapping and task speed scheduling simultaneously. Experimental results show that the proposed algorithms can achieve significant energy savings (e.g., 24% on average for 70 nm technology) over the baseline that only considers the response time requirement.