QRQW PRAM: Accounting for contention in parallel algorithms

Phillip B. Gibbons*, Yossi Matias, Vijaya Ramachandran

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

41 Scopus citations

Abstract

We introduce the queue-read, queue-write (QRQW) PRAM model, which permits concurrent reading and writing, but at a cost proportional to the number of readers/writers to a memory location in a given step. Previously, there were no formal complexity models that accounted for the contention to memory locations, despite the large impact of such contentions on the performance of parallel programs. The QRQW PRAM is strictly more powerful than the EREW PRAM. We show a separation of √lg n between the two models, and present faster and more efficient QRQW algorithms for many basic problems. Nevertheless, we show that the QRQW can be efficiently emulated with only logarithmic slowdown on Valiant's BSP model, and hence on hypercube-type non-combining networks, even when latency, synchronization, and memory granularity overheads are taken into account. In contrast, efficient emulations for the CRCW PRAM on such networks are only known with polynomial slowdown. Most existing machines obey the queue-read, queue-write rule. We study the impact of the QRQW rule on algorithm design, devising new techniques for low-contention algorithms. Our results include fast and efficient algorithms for computing the OR, leader election, linear compaction, multiple compaction, integer sorting, and CRCW simulation, as well as several lower bounds. Some of the results presented are quite involved and use several novel ideas and techniques that are interesting on their own.

Original languageEnglish
Title of host publicationProceedings of the Annual ACM SIAM Symposium on Discrete Algorithms
PublisherAssociation for Computing Machinery (ACM)
Pages638-648
Number of pages11
ISBN (Print)0898713293
StatePublished - 1994
Externally publishedYes
EventProceedings of the Fifth Annual SIAM Symposium on Discrete Algorithms - Arlington, VA, USA
Duration: 23 Jan 199425 Jan 1994

Publication series

NameProceedings of the Annual ACM SIAM Symposium on Discrete Algorithms

Conference

ConferenceProceedings of the Fifth Annual SIAM Symposium on Discrete Algorithms
CityArlington, VA, USA
Period23/01/9425/01/94

Fingerprint

Dive into the research topics of 'QRQW PRAM: Accounting for contention in parallel algorithms'. Together they form a unique fingerprint.

Cite this