Clustering algorithm for parallelizing software systems inmultiprocessors environment

摘要

A variety of techniques and tools exist to parallelize softwarensystems on different parallel architectures (SIMD, MIMD). With thenadvances in high-speed networks, there has been a dramatic increase innthe number of client/server applications. A variety of client/servernapplications are deployed today, ranging from simple telnet sessions toncomplex electronic commerce transactions. Industry standard protocols,nlike Secure Socket Layer (SSL), Secure Electronic Transaction (SET),netc., are in use for ensuring privacy and integrity of data, as well asnfor authenticating the sender and the receiver during message passing.nConsequently, a majority of applications using parallel processingntechniques are becoming synchronization-centric, i.e., for every messagentransfer, the sender and receiver must synchronize. However, moreneffective techniques and tools are needed to automate the clustering ofnsuch synchronization-centric applications to extract parallelism. Thenauthors present a new clustering algorithm to facilitate thenparallelization of software systems in a multiprocessor environment. Thennew clustering algorithm achieves traditional clustering objectivesn(reduction in parallel execution time, communication cost, etc.).nAdditionally, our approach: 1) reduces the performance degradationncaused by synchronizations, and 2) avoids deadlocks during clustering.nThe effectiveness of our approach is depicted with the help ofnsimulation results