Design and Analysis of Parallel Content Matching Algorithm for Pub-Sub Systems using Different Parallel Paradigms

The key challenge in the performance of Pub-Sub system is the design of matching algorithm. The content matching takes place on each broker system along the path from publisher to subscriber in broker overlay network. Matching time is significant as compared with a network delay of message forwarding. In the case of content-based systems, matching is time-consuming task, whose performance affects the entire system. Efficient content-based event matching is considered as challenging research problem from past few years. All algorithms proposed earlier are inherently sequential and does not exploit parallel architecture which is easily available in current generation computers. This paper describes a new Pub-Sub content-based matching algorithm designed using principles of shared and distributed memory program running efficiently on multicore processor architecture. Hybrid parallel programming approach shows 4 times reduction in average matching time and an improved throughput of over 4000 events/s when using 32 processors which are almost double of events processed using only shared memory approach or only distributed memory approach. Paper also presents the result of the content matching algorithm using GPU.