An Improved Lazy Release Consistency Model

Chapram Sudhakar,T Ramesh

doi:10.3844/jcssp.2009.778.782

Abstract

Problem statement: A network of workstations, viewed as a distributed shared memory system can be used to develop and test parallel algorithms. Approach: For implementing parallel algorithms on such DSMs shared memory consistency model plays a vital role. Results: However on a LAN, strict consistency models like Sequential Consistency model (SC) are not useful since the communication is slow. In such environments relaxed models like Entry Consistency (EC), Release Consistency (RC) or their variations such as Lazy Release Consistency (LRC) are generally used. Conclusion/Recommendations: In this study an Improved Lazy Release Consistency (ILRC) model is proposed. This model is studied with standard parallel algorithms. In many cases the ILRC model is proved to work better than the LRC model.

Highlights

Usage of network of workstations for parallel processing is very common
There is a variation of Release Consistency model known as Lazy Release Consistency (LRC) model which is used in TreadMarks[4] system
Lazy Release Consistency Model ensures that all programs without data races behave as if they were executing on a conventional Sequentially Consistent (SC) memory

Summary

Introduction

Usage of network of workstations for parallel processing is very common Such an environment can be viewed by the programmer as a message passing environment or Distributed Shared Memory (DSM)[7] environment. There is a variation of Release Consistency model known as Lazy Release Consistency (LRC) model which is used in TreadMarks[4] system. This model works better for parallel algorithms which manipulate small set of data items in a brief critical section that results in very little modifications in the page of those data items. Lazy Release Consistency Model ensures that all programs without data races behave as if they were executing on a conventional Sequentially Consistent (SC) memory.

Methods

Results

Conclusion