Replicating the R in URL

The Web supports access to a very wide variety of services and objects via the simple naming mechanism provided by the Uniform Resource Locator or URL. When resources are read-only entities, which is still the common case, there is little problem in returning their output to clients in the face of multiple concurrent accesses. However, the Web is increasingly being adopted as the front-end for distributed multi-user applications, in which case resources must be maintained in an environment characterised by multiple concurrent reads and writes, and dynamic access control. This engenders basic needs for concurrency control and dynamic configuration in the single server scenario. Highly interactive Web environments, which do not benefit from caching, cause sufficient extra server load to necessitate the use of multiple servers. Replication is necessary when caching fails. The moderate complexity of single server resource management is then compounded. This report details a design for resource management that supports arbitrary resource types, replicated instances of a given resource type, and multiple co-operating servers. It is based on our experiences of coping with complexity, in the context of distributed learning environments