Workload Type-Aware Scheduling on big.LITTLE Platforms

A4 Conference proceedings


Internal Authors/Editors


Publication Details

List of Authors: Simon Holmbacka, Jörg Keller
Publication year: 2017
Publisher: Springer
Book title: Algorithms and Architectures for Parallel Processing. ICA3PP 2017
Title of series: Lecture Notes in Computer Science
Volume number: 10393
Start page: 3
End page: 17
ISBN: 978-3-319-65481-2
eISBN: 978-3-319-65482-9
ISSN: 0302-9743


Abstract

Abstract Optimizing energy efficiency in execution strategies has tra-
ditionally been heavily influenced by hardware mechanisms such as fre-
quency scaling and core sleep states. With such facilities, the system
can be scaled dynamically and on-demand to trade power dissipation
for clock speed or parallelism. Determining the most efficient execution
configuration has been described in much related work, but few efforts
have been put on including the workload type into the calculation. The
type of the workload affects both the performance and the power of the
processor, and is especially important when considering heterogeneous
systems like the big.LITTLE, since different cores handle the workload
with different efficiency. In this paper, we demonstrate the influence of
the workload type when choosing an optimal execution strategy on a
big.LITTLE platform. We implement schedulers capable of including
workload type, and we provide a runtime system capable of executing the
schedules on a real-world platform. Results demonstrate that including
workload types into the scheduler saves between 7.1% and 31.3% of energy
in our best/worst corner case studies, a result that should be considered
in future implementations of big.LITTLE schedulers.


Last updated on 2019-15-10 at 03:07