SPLMax: Exploiting the Simple Path Introduced Locality for Maximum Flow Acceleration

As a basic problem in graph theory, the maximum flow (max-flow) problem has important applications in networking and communication related areas. The simple path introduced locality is implicit in classic max-flow algorithms, i.e., only the vertices in simple paths between source and sink are involv...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE communications letters Jg. 22; H. 7; S. 1330 - 1333
Hauptverfasser: Wei, Wei, Liu, Yang, Zhang, Ruqing
Format: Journal Article
Sprache:Englisch
Veröffentlicht: IEEE 01.07.2018
Schlagworte:
Acceleration
Buildings
Computational modeling
cut vertex
graph preprocessing
Graph theory
Indexes
Maximum flow
overlay graph
simple path
Time complexity
ISSN:1089-7798, 1558-2558
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:As a basic problem in graph theory, the maximum flow (max-flow) problem has important applications in networking and communication related areas. The simple path introduced locality is implicit in classic max-flow algorithms, i.e., only the vertices in simple paths between source and sink are involved in max-flow calculation. However, this kind of locality is completely ignored in existing acceleration methods, which leads to a lot of useless calculations and seriously degrades the acceleration effect. We propose simple-path locality based max-flow acceleration algorithm (SPLMax) to address the problem, where an overlay graph is built and used to accelerate calculation by only including necessary vertices. Random graph based simulations show that with SPLMax, at best only 0.001% vertices (i.e., 1/71193) in the graph need to be involved in max-flow calculation. For the comparison using real-world graphs, SPLMax has the minimal pre-processing time (at most 109 times faster than other methods) and minimal average max-flow computation time (at most 4.3 times faster than other methods).
ISSN:1089-7798
1558-2558
DOI:10.1109/LCOMM.2018.2830786