BibTeX
@MISC{_distrm:distributed,
author = {},
title = {DistRM: Distributed Resource Management for On-Chip Many-Core Systems},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
{lohmann, wosch} @ cs.fau.de The trend towards many-core systems comes with various is-sues, among them their highly dynamic and non-predictable workloads. Hence, new paradigms for managing resources of many-core systems are of paramount importance. The problem of resource management, e.g. mapping applications to proces-sor cores, is NP-hard though, requiring heuristics especially when performed online. In this paper, we therefore present a novel resource-management scheme that supports so-called malleable applications. These applications can adopt their lev-el of parallelism to the assigned resources. By design, our (de-centralized) scheme is scalable and it copes with the computa-tional complexity by focusing on local decision-making. Our simulations show that the quality of the mapping decisions of our approach is able to stay near the mapping quality of state-of-the-art (i.e. centralized) online schemes for malleable appli-cations but at a reduced overall communication overhead (only about 12,75 % on a 1024 core system with a total workload of 32 multi-threaded applications). In addition, our approach is scalable as opposed to a centralized scheme and therefore it is practically useful for employment in large many-core systems as our extensive studies and experiments show.
Keyphrases
resource management on-chip many-core system malleable appli-cations performed online core system local decision-making computa-tional complexity multi-threaded application many-core system mapping decision mapping quality extensive study non-predictable workload new paradigm online scheme novel resource-management scheme large many-core system centralized scheme total workload paramount importance assigned resource so-called malleable application various is-sues proces-sor core mapping application reduced overall communication overhead
