A Modified Approach to Improve the Performance of Lazy Release Consistency (LRC) Model

Abstract

This paper proposes a method for improving the performance of existing Lazy Release Consistency (LRC) model. In LRC model,
before accessing a shared page a processor must acquire a lock associated with it as well as achieve all modifications made by other
processors on the same shared page from the last lock releaser. All modifications are maintained by a data structure known as diff. For
each access to a shared page a new diff is being created. As a result, the number of total diffs is increased in parallel with the access to
shared page, causing a large amount of message transfer at the time of acquiring a lock. The proposed technique reduces the network
traffic at a great extent by transferring only the critical section of shared page instead o f transferring a large number of diffs. It also
ensures the transmission of message in reduced size for a particular page. Moreover, by eliminating the need of maintaining the
redundant copies of a diff, the proposed method also reduces the vast memory requirement of LRC model. From the experimental
results, it is found that the proposed modification significantly minimizes the message transfer as well as the memory requirement
without affecting the functionalities of the LRC model.