Social network ordering based on communities to reduce cache misses

Abstract : ABSTRACT. One of social graph's properties is the community structure, that is, subsets where nodes belonging to the same subset have a higher link density between themselves and a low link density with nodes belonging to external subsets. Futhermore, most social network mining algorithms comprise a local exploration of the underlying graph, which consists in referencing nodes in the neighborhood of a particular node. The idea of this paper is to use the community structure during the storage of large graphs that arise in social network mining. The goal is to reduce cache misses and consequently, execution time. After formalizing the problem of social network ordering as a problem of optimal linear arrangement which is known as NP-Complet, we propose NumBaCo, a heuristic based on the community structure. We present for Katz score and Pagerank, simulations that compare classic data structures Bloc and Yale to their corresponding versions that use NumBaCo. Results on a 32 cores NUMA machine using amazon, dblp and web-google datasets show that NumBaCo allows to reduce from 62% to 80% of cache misses and from 15% to 50% of execution time.
Document type :
Journal articles
Complete list of metadatas

Cited literature [16 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01304968
Contributor : Thomas Messi Nguélé <>
Submitted on : Wednesday, May 10, 2017 - 2:46:41 PM
Last modification on : Friday, May 24, 2019 - 4:10:17 PM
Long-term archiving on : Friday, August 11, 2017 - 1:13:47 PM

File

ARIMA24pp23-47_1.pdf
Files produced by the author(s)

Licence


Public Domain

Identifiers

  • HAL Id : hal-01304968, version 5

Relations

  • is version of hal-01304968 - version mise à jour à la demande de l'editeur d'arima

Citation

Thomas Messi Nguélé, Maurice Tchuente, Jean-François Méhaut. Social network ordering based on communities to reduce cache misses. Revue Africaine de la Recherche en Informatique et Mathématiques Appliquées, INRIA, 2017, Special issue CRI 2015, Volume 24 - 2016-2017 - Special issue CRI 2015, pp.23 - 47. ⟨hal-01304968v5⟩

Share

Metrics

Record views

425

Files downloads

518