NINOS Take Five: The Management Infrastructure for Distributed Event-Driven Workflows

摘要

Many workflows are inherently distributed over large-scale enterprise networks. These workflows involve many collaborating partners that interact in an autonomous and event-driven manner. Managing these workflows in an efficient and reliable manner is a challenging task. In this paper, we propose and evaluate the design for a distributed workflow management system that runs according to a set of protocols which utilize a content-based publish/subscribe messaging substrate. The novel features of our framework include elastic management of resources and flexible re-location of management entities to ensure cost-effective and low-latency management operations. Our experimental evaluation shows that the distributed and event-driven approach scales and maintains constant response time for varying management workloads. The average response time for management operations was reduced by 25% using our elastic management cluster approach compared to a fixed management cluster. Management requests were processed up to 10 times faster.