Orchestrating computations on the world-wide web

Abstract

Word processing software, email, and spreadsheet have revolutionized office activities. There are many other office tasks that are amenable to automation, such as: scheduling a visit by an external visitor, arranging a meeting, and handling student application and admission to a university. Many business applications —protocol for filling an order from a customer, for instance— have similar structure. These seemingly trivial examples embody the computational patterns that are inherent in a large number of applications, of coordinating tasks at different machines. Each of these applications typically includes invoking remote objects, calculating with the values obtained, and communicating the results to other applications. This domain is far less understood than building a function library for spreadsheet applications, because of the inherent concurrency. We address the task coordination problem by (1) limiting the model of computation to tree structured concurrency, and (2) assuming that there is an environment that supports access to remote objects. The environment consists of distributed objects and it provides facilities for remote method invocation, persistent storage, and computation using standard function library. Then the task coordination problem may be viewed as orchestrating a computation by invoking the appropriate methods in proper sequence. Tree structured concurrency permits only restricted communications among the processes: a process may spawn children processes and all communications are between parents and their children. Such structured communications, though less powerful than interactions in process networks, are sufficient to solve many problems of interest, and they avoid many of the problems associated with general concurrency.