A Method to Construct Task Scheduling Algorithms for Heterogeneous Multi-Core Systems

Abstract

The use of heterogeneous multicore processors (HMP) is spreading rapidly from data centers to large-scale deployment in smartphones because they give greater flexibility to adapt to power constraints and performance needs. In this study, we show that an intelligent task scheduler is critical for improving the performance and energy efficiency in an HMP environment. We assume that the tasks are independent in the environment, with hard real-time constraints and multicore systems, where the processors can be manipulated to change the clock cycle speed and power levels. Tasks are assumed to arrive aperiodically where the tasks are applications from the SPEC CPU 2006 benchmark suite. In the evaluation, we used a real system comprising of two multicore processors, which supported on-the-fly dynamic voltage/frequency scaling. We extracted several important components from previously proposed algorithms and combined them to construct algorithms with better performance. Our results showed that some of the best combinations reduced the energy consumption and achieved a better completion rate in the environment. In addition, a method is proposed for calculating the upper-bound of the task completion rate and energy consumption so that there is a guide as to how near the results are to the optimal performance. INDEX TERMS Task scheduling, energy-aware scheduling, heterogeneous multi-core, real-time scheduling, dynamic voltage/frequency scaling.

Flexible hardware approach to multi‐core time‐predictable systems design based on the interleaved pipeline processing

“…Task scheduling process belongs to one of the most important procedures during the designing of the real-time applications as it decides about appropriate and optimal resource utilisation and overall quality of the system [2,[26][27][28][29][30][31][32]. That is why this phase of the contemporary real-time system manufacturing belongs to one of the most crucial issues.…”

“…Tasks assignment to resources depends on the processing speed. There are two main approaches: Short task to the Fast processor, Long task to the Slow processor (SFLS) or Short task to the Slow processor and Long task to the Fast processor (SSLF) [32]. A detailed analysis of the expected execution time of the tasks, and its sensitivity to the surroundings (system load, tasks frequency, number of cores, memory systems, constraints etc.)…”

“…We can distinguish a lot of techniques and algorithms based on various factors and parameters reflecting properties of a given system or a platform. Techniques operating on a network of heterogeneous processors [26, 29, 30, 32] differentiate the processing cores efficiency. Tasks assignment to resources depends on the processing speed.…”

“…[27, 28, 31] is very important. The results of such analyses lead to the formulation of the tasks' schedulability conditions [2, 32], i.e. determining the system properties that allow appropriate tasks scheduling and meeting timing constraints.…”

A Novel Fast Parallel Batch Scheduling Algorithm for Solving the Independent Job Problem

“…The purpose of this problem is to allot each job to a batch and to schedule the batch on the machine to minimize the maximum completion time of the schedule, C max = max j C j , where C j represents the completion time of job j in the schedule [10,11]. Using the notations proposed in [12,13], this problem can be denoted as P s j , p − batch, K i C max .…”