Search

Models for energy-efficient static scheduling of parallelizable tasks with deadlines on frequency-scalable parallel machines comprise moldable vs. malleable tasks and continuous vs. discrete frequency levels, plus preemptive vs. non-preemptive task execution with or without task migration. We investigate the tradeoff between scheduling time and energy efficiency when going from continuous to discrete core allocation and frequency levels on a multicore processor, and from preemptive to non-preemptive task execution. To this end, we present a tool to convert a schedule computed for malleable tasks on machines with continuous frequency scaling [Sanders and Speck, Euro-Par (2012)] into one for moldable tasks on a machine with discrete frequency levels. We compare the energy efficiency of the converted schedule to the energy consumed by a schedule produced by the integrated crown scheduler [Melot et al., ACM TACO (2015)] for moldable tasks and a machine with discrete frequency levels. Our experiments with synthetic and application-based task sets indicate that the converted Sanders Speck schedules, while computed faster, consume more energy on average than crown schedules. Surprisingly, it is not the step from malleable to moldable tasks that is responsible but the step from continuous to discrete frequency levels. One-time frequency scaling during a task's execution can compensate for most of the energy overhead caused by frequency discretization. [ABSTRACT FROM AUTHOR]

In this paper, we focus on the problem of scheduling a collection of similar task graphs on a heterogeneous platform, when the task graph is an intree. We rely on steady-state scheduling techniques, and aim at optimizing the throughput of the system. Contrarily to previous studies, we concentrate on practical aspects of steady-state scheduling, when dealing with a collection (or batch) of limited size. We focus here on two optimizations. The first one consists in reducing the processing time of each task graph, thus making steady-state scheduling applicable to smaller batches. The second one consists in degrading a little the optimal-throughput solution to get a simpler solution, more efficient on small batches. We present our optimizations in details, and show that they both help to overcome the limitation of steady-state scheduling: our simulations show that we are able to reach a better efficiency on small batches, to reduce the size of the buffers, and to significantly decrease the processing time of a single task graph (latency). [ABSTRACT FROM AUTHOR]