Multiprocessor Global Scheduling on Frame-Based DVFS Systems

In this ongoing work, we are interested in multiprocessor energy efficient systems, where task durations are not known in advance, but are know stochastically. More precisely, we consider global scheduling algorithms for frame-based multiprocessor stochastic DVFS (Dynamic Voltage and Frequency Scaling) systems. Moreover, we consider processors with a discrete set of available frequencies.

💡 Research Summary

The paper addresses the challenge of energy‑efficient scheduling on multiprocessor platforms where task execution times are not known a priori but follow stochastic distributions. The authors adopt a frame‑based model: a fixed‑length time window (the frame) is defined, and all jobs arriving at the beginning of a frame must finish before the frame ends. This model simplifies synchronization across cores and provides a natural boundary for global scheduling decisions. Each task τi is characterized by a probability distribution (e.g., normal, log‑normal) with mean μi and variance σi², obtained through profiling or online measurement. The processors support a discrete set of DVFS levels F = {f1, f2, …, fk}. Every level fj is associated with a voltage Vj, power consumption Pj, and an execution speed sj (cycles per second). Switching between levels incurs a transition cost Ctrans(j→j′), which the algorithm explicitly accounts for.
The core contribution is a two‑stage global scheduler. In the first stage, the algorithm determines a mapping of tasks to cores. It defines a cost function C(i,c) = α·E