By Topic

Moving home-based lazy release consistency for shared virtual memory systems

Sign In

Cookies must be enabled to login.After enabling cookies , please use refresh or reload or ctrl+f5 on the browser for the login options.

Formats Non-Member Member
$31 $13
Learn how you can qualify for the best price for this item!
Become an IEEE Member or Subscribe to
IEEE Xplore for exclusive pricing!
close button

puzzle piece

IEEE membership options for an individual and IEEE Xplore subscriptions for an organization offer the most affordable access to essential journal articles, conference papers, standards, eBooks, and eLearning courses.

Learn more about:

IEEE membership

IEEE Xplore subscriptions

4 Author(s)
Jae Woong Chung ; Central Res. Inst., Samsung Heavy Ind. Co. Ltd., Taejon, South Korea ; Byeong Hag Seong ; Kyu Ho Park ; Daeyeon Park

While Shared Virtual Memory is a cost-effective way to provide an illusion of shared memory to programmers, the performance of the system is limited by its inherent overheads. Several protocols such as Lazy Release Consistency (LRC) and Home-based LRC (HLRC) have been suggested to overcome the overheads. In this paper we propose a new all-software protocol, Moving Home-based LRC (MHLRC). Its main contribution is to introduce a new concept, Moving Home. Using it, our protocol provides 4 efficient functionalities: (a) providing migration mechanism, (b) eliminating the duty of updating the home copy at the end of an interval, (c) providing an adaptive home locating scheme, and (d) simplifying the update procedure on a stale copy. To evaluate our protocol, we implemented a prototype of MHLRC. The experiment with 4 benchmarks showed that our protocol improved the performance of the system by on the average 10% over LRC, and by up to 96% over HLRC. From the detailed analysis, we found that the improvement mainly comes from the reduction of the coherence action overhead. Our protocol reduced the number of diff operations executed during computation up to 64%(FFT)~0.4%(SOR) against LRC and 56%(FFT)~0.3%(SOR) against HLRC. It also reduces the requirement for memory and network bandwidth

Published in:

Parallel Processing, 1999. Proceedings. 1999 International Conference on

Date of Conference: