A Decentralized Load Balancing Approach for Parallel Search-Tree Optimization

Current generation supercomputers have over one million cores awaiting highly demanding computations and applications. An area that could largely benefit from such processing capabilities is naturally that of exact algorithms for NP-hard problems. We propose a general implementation framework that t...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Proceedings (International Symposium on Parallel and Distributed Computing) s. 173 - 178
Hlavní autoři: Abu-Khzam, F. N., Mouawad, A. E.
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 01.12.2012
Témata:
Dominating Set
Dynamic Load Balancing
Exact Algorithms
Heuristic algorithms
Load management
Network topology
Optimization
Parallel Algorithms
Recursive Backtracking
Scalability
Supercomputers
Topology
Vertex Cover
ISSN:2379-5352
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Current generation supercomputers have over one million cores awaiting highly demanding computations and applications. An area that could largely benefit from such processing capabilities is naturally that of exact algorithms for NP-hard problems. We propose a general implementation framework that targets highly scalable parallel exact algorithms for NP-hard graph problems. We tackle the problems of efficiency and scalability by combining a fully decentralized dynamic load balancing strategy with special implementation techniques for exact graph algorithms. As a case-study, we use our framework to implement parallel algorithms for the VERTEX COVER and DOMINATING SET problems. We present experimental results that show notable improved running times on all types of input instances.
ISSN:2379-5352
DOI:10.1109/PDCAT.2012.16