[Re] Speedup Graph Processing by Graph Ordering - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue The ReScience journal Année : 2021

[Re] Speedup Graph Processing by Graph Ordering

Fabrice Lécuyer
Maximilien Danisch
  • Fonction : Auteur
  • PersonId : 940804
  • IdRef : 188447407
Lionel Tabourier
  • Fonction : Auteur

Résumé

Cache systems keep data close to the processor to access it faster than main memory would. Graph algorithms benefit from this when a cache line contains highly related nodes. Hao Wei extitet al. propose to reorder the nodes of a graph to optimise the proximity of nodes on a cache line. Their contribution, Gorder, creates such an ordering with a greedy procedure. In this replication, we implement ten different orderings and measure the execution time of nine standard graph algorithms on nine real-world datasets. We monitor cache performances to show that runtime variations are caused by cache management. We confirm that Gorder leads to the fastest execution in most cases due to cache-miss reductions. Our results show that simpler procedures are yet almost as efficient and much quicker to compute. This replication validates the initial results but highlights that generating a complex ordering like Gorder is time-consuming. A replication of [1]. 2 Method The original study [1] was motivated by the observation that cache stall can take up to 70% of the whole computation time, which is supported by the observations reported
Fichier principal
Vignette du fichier
article.pdf (638.56 Ko) Télécharger le fichier
Origine : Fichiers éditeurs autorisés sur une archive ouverte

Dates et versions

hal-03261251 , version 1 (15-06-2021)

Identifiants

Citer

Fabrice Lécuyer, Maximilien Danisch, Lionel Tabourier. [Re] Speedup Graph Processing by Graph Ordering. The ReScience journal, 2021, 7 (1), pp.#3. ⟨10.5281/zenodo.4836230⟩. ⟨hal-03261251⟩
101 Consultations
139 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More