A Performance Evaluation of Distributed Algorithms on Shared Memory and Message Passing Middleware Platforms
Sanjay Ahuja, Roger E. Eggen, Anjani K. Jha · 2005
The fundamental characteristics of a distributed computing environment are heterogeneity, partial failure, latency and difficulty of “gluing together ” multiple, independent processes into a robust, scalable application. JavaSpaces, which is a shared memory paradigm, provides high-level coordination mechanism for Java easing the burden of creating distributed applications. A large class of distributed problems can be approached using JavaSpaces simple framework. JavaSpaces allows processes to communicate even if each was wholly ignorant of the others. Common Object Request Broker Architecture (CORBA), on the other hand, is a standard developed by the Object Management Group (OMG), which allows communication between objects that are written in different programming languages. It provides common message passing mechanism for interchanging data and discovering services. In this project, we compare these two platforms for distributed computing both quantitatively and qualitatively. To do so, we analyze the performance of distributed algorithms that divide a task into small sub-tasks which are distributed over a network of computers to perform computations in parallel. Specifically, we measure the performance of an insertion sort algorithm of O (n2) complexity on both the JavaSpaces and CORBA platforms. We measure latency, speed-up, and efficiency and analyze the implications on overall performance and scalability. Povzetek: Članek opisuje ovrednotenje porazdeljenih algoritmov na platformah. 1