On Compact Mappings for Multicore Systems
Application mapping is key for efficient multicore processing, i.e., selecting which resources to allocate to a given application, like computation to cores. Mapping is increasingly difficult in multi-application scenarios, where resource contention might degrade the performance of an application. I...
Saved in:
Published in | Embedded Computer Systems: Architectures, Modeling, and Simulation pp. 325 - 335 |
---|---|
Main Authors | , , |
Format | Book Chapter |
Language | English |
Published |
Cham
Springer International Publishing
|
Series | Lecture Notes in Computer Science |
Online Access | Get full text |
Cover
Loading…
Summary: | Application mapping is key for efficient multicore processing, i.e., selecting which resources to allocate to a given application, like computation to cores. Mapping is increasingly difficult in multi-application scenarios, where resource contention might degrade the performance of an application. In order to solve this, a promising avenue is to consider “compact” mappings, those which require a small and (geometrically) compact area within the chip. Compact mappings should decrease contention between applications by providing regional isolation and allowing multiple applications to be mapped simply. Previous work has shown that compact mappings can significantly outperform mappings obtained with a random strategy. In this paper we investigate the promise of compact mappings by running extensive simulations on Noxim, a cycle-accurate network-on-chip simulator. Results show the promises of compact mappings do not hold up in practice. When comparing to mappings selected with a heuristic better than simply choosing cores at random, our experiments do not indicate significant advantages from compact mappings. We outline possible reasons for this. |
---|---|
ISBN: | 9783030275617 3030275612 |
ISSN: | 0302-9743 1611-3349 |
DOI: | 10.1007/978-3-030-27562-4_23 |