A Comparison of RPC and Mobile Agent Paradigm for Scalable Workflow Systems Using Petri Nets

摘要

Until now RPC paradigm has been adopted as a communication facility within most workflow systems. As the number of workflow process increases, RPC-based workflow systems suffer from performance degradation. To address this limited scalability, mobile agents have been considered as an alternative to the RPC-based architecture recently. However, since a mobile agent contains the entire workflow definition, the physical size is apt to be much larger than a simple RPC message. Consequently, the migration of mobile agents over the network causes another communication overhead. In this paper, we propose a workflow system model based on mobile agents, so called Maximal Sequence model, as an alternative to conventional RPC-based and the previous mobile agent-based models. The proposed model segments a workflow definition into blocks, and assigning each of them to a mobile agent. We also construct three stochastic Petri net models of conventional RPC-based, the previous mobile agent-based (DartFlow), and the Maximal Sequence model-based workflow systems to compare performance and scalability. The stochastic Petri-net simulation results show that the proposed model outperforms the previous ones as well as comes up with better scalability when the numbers of workflow tasks and concurrent workflows are relatively large.