GDP: A Paradigm for Intertask Communication in Grid Computing Through Distributed Pipes

Existing grid models target purely data parallel applications without inter-task communication. This paper proposes a transparent programming model to support communicating parallel tasks in a wide area grid. The proposed grid model with Distributed Pipes (DP) abstraction named as, GDP, enables location independent intertask communication among processes on machines spread over a wide area distributed system. This approach enables anonymous migration of communicating parallel tasks adjusting to grid dynamics. The proposed model supports sequential load to coexist with parallel load. A prototype of the proposed model has been implemented over clusters of nodes spread across the Internet. A steady state equilibrium engineering problem was studied over the model. Performance studies show linear to super linear speed up for the application.