Optimal and Robust Routing of Subscriptions for Unifying Access to the Past and the Future in Publish/Subscribe.

摘要

A flexible, scalable, and asynchronous middleware abstract is needed for business process management, which involves thousands of tasks and a large number of running instances of large business processes. The content-based publish/subscribe system is an ideal candidate to serve as enterprise service bus for these applications. In the publish/subscribe paradigm, information providers called publishers disseminate publications to all subscribers who have expressed interests by registering subscriptions through a loosely coupled interface. However, the traditional publish/subscribe paradigm only supports "stateless" subscriptions, that is, event correlation is ignored. Moreover, subscribers can only receive publications issued after their subscriptions. There are many application contexts, however, where access to publications from the past is necessary, such as for replaying a business process execution to debug it. Even more interesting uses arise when data from the past can be correlated with those in the future. Therefore, new languages and new functionalities are needed in the standard publish/subscribe model in order to support business process management.;With the above features, we propose the NINOS system, the distributed business process execution architecture as a case study, which uses light-weight activity agents to carry out business process execution in a distributed environment. NINOS proves that decentralized business process execution is the trend for next generation products, and the publish/subscribe model is ideal to serve as an enterpriser service bus (ESB) for distributed applications.;A new subscription language PADRES SQL(PSQL) which can express event patterns and unify both historic and future views for subscribers. PADRES allows a subscriber to access data published both in the past and in the future. Furthermore, complex event detection happens in the broker network. The main difficulties of distributed event detection are routing a composite subscription, including where and how to decompose the composite subscription, and routing the individual parts of the subscription. Our composite subscription routing decisions are based on a cost model which minimizes the routing and detection delay. An adaptive subscription routing protocol is proposed to determine efficient location with dynamic changing workloads. PADRES also provides robust message delivery by exploring alternative paths in a cyclic overlay. Routing optimizations and efficient matching algorithms are studied to improve the performance of the extended publish/subscribe model.