DISTRIBUTED COORDINATION MODELS FOR CLIENT-SERVER COMPUTING

Interactions between distributed applications presuppose an underlying control model to coordinate information exchanges and networking soft ware to implement that model. The client/server control model defines distributed interactions in terms of one program requesting and obtain ing a service from a second, possibly remote, application. However, th is basic model provides inadequate design support when clients need to invoke multiple, independent services, coordinated to reflect how tho se services interrelate and contribute to the overall application. The author describes extensions to the basic client/server model that exp licitly address one-to-many client/server interactions by discussing t hree basic design issues: how multiple services are requested, how tho se services are managed, and how clients receive responses. The extend ed coordination models provide support for locating, obtaining, and sy nchronizing services, as well as for collecting and combining results from multiple servers in a manner that is transparent to clients. Exte nded models include a scripting engine for managing data and temporal dependencies among services; a basic request broker for mediating clie nt access to distributed services; and extended request broker models that decompose composite services, manage redundant servers, and repli cate messages to logical server groups. These coordination models were designed as generic, programmable control services. The control servi ces are interoperable, so they can be combined like building blocks to match application-specific coordination requirements. The one-to-many coordination services are layered on top of an object-oriented, messa ge-passing communication substrate, which transparently manages the co mplexities of interprogram interactions across networks of heterogeneo us computers. This layered architecture lets complex coordination beha viors be modeled and executed external to application elements. It acc omplishes this through high-level application programming interface (A PI) calls. The resulting partitioning of application and generic distr ibuted behaviors yields improved modularity, maintainability, and exte nsibility of individual clients and servers.