Programmability and performance of M++ self-migrating threads

Fukuda, Munehiro; Suzuki, N.; Campos, Luis Miguel; Kobayashi, Satoshi

doi:10.1109/clustr.2001.959994

Cited by 4 publications

(1 citation statement)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since a new thread spawns along each edge, a large graph would results in considerable thread migration, creation, and termination overheads. The authors previously developed M++ self-migrating threads [11] that were instantiated at the user level, thus not involving any operating system interventions, and capable of migrating over a distributed graph. Despite of its performance tune-up, the system still suffered from thread management overheads.…”

Section: B Parallelization Approachesmentioning

confidence: 99%

Agent and Spatial Based Parallelization of Biological Network Motif Search

Kipps

Kim

Fukuda

2015

2015 IEEE 17th International Conference on High Performance Computing and Communications, 2015 IEEE 7th International Symposium

View full text Add to dashboard Cite

Most graph algorithms are challenging in parallelization, in particular executing fine-grain computation at each graph node in parallel from both programmability and performance viewpoints. To bridge the semantic gap between the original sequential algorithms and their corresponding parallelized programs, we have been developing MASS: a parallel library for multi-agent spatial simulation. The library allows software agents to crawl a distributed array, e.g., a graph mapped over a cluster system. To demonstrate the MASS library's fitness to graph parallelization, we have focused on biological network motif search. This paper compares three different parallelizing approaches such as the MASS agent-based, MASS array-based, and the conventional MPI parallelizations, and discusses the MASS library's applicability to graph algorithms.

show abstract

Section: B Parallelization Approachesmentioning

confidence: 99%