Coherence Models for Web Object-Based Systems

In order to overcome the high latencies involved with accessing remote Web objects, caching is often used. Unfortunately there are a number of problems with current caching mechanisms. First of all, the client/server model used in the Web does not scale well. Second, the replication and coherence strategies used are implemented at the client side according to a \one policy ts all" principle and are applied indiscriminately to all Web objects. With a growing variety of Web objects, however, such general-purpose solutions are rapidly becoming inadequate. Finally, current replication strategies are deened by implementations while formalized underlying coherence models are missing. As new applications with new requirements are brought to the Internet, the current solutions will not be capable of evolving to accommodate them. This paper proposes an approach to exible object-based replication and coherence support for the Web. Our approach is based on a symmetric model of distribution, as opposed to the asymmetric client/server model. We take into account the diversity of Web objects by applying a replication and a coherence strategy on a per-object basis. We also introduce a two-level coherence model which allows replication and coherence to be based on both server knowledge and client preferences. Server-based models are applied by object developers and client-based models allow a client to specify required consistency guarantees 7]. The goal of our proposal is to provide a scalable and exible Web infrastructure that is able to support a wide range of coherence and replication strategies. Our model is based on the use of the Globe architecture 3]. In Globe, processes interact and communicate though distributed shared objects. Each object ooers one or more interfaces, each consisting of a set a methods. Distributed shared objects are physically distributed and all implementation aspects including replication strategies are part of the object and hidden behind its interface. A distributed shared object is built from local objects. As soon as a replica is created, it becomes a part of the distributed shared object with no distinction between it and the other parts. Thus, a Web application is represented as one or several distributed shared objects where both the server and cached data are represented by local objects. The main advantage of such a model is that we can implement object-speciic policies.