The semantic web promises distributed processing of knowledge structures such as RDF and Topic Maps in an intelligent fashion. However, while lots of effort has been put into defining these representations very little has been defined in terms of how clients and servers, or semantic web peers, will actually communicate, what they will be able to say and what happens when messages from clients arrive at servers.

The problem can be better expressed as being a problem of how to interact with a Semantic Web Server. At the moment the semantic web is not a web but multiple islands of semantic information. This information may be available from a web server in a serialized form but there is no standardized way, and only a few non-standardized ways, to query and update remote RDF or Topic Map models.

This paper introduces a conceptual framework that identifies several key areas that must be addressed in defining a distributed knowledge processing environment. These key areas can be summarized as the structures to be interchanged, the abstract protocol that defines what messages are available for clients to invoke, how these protocols are bound into the real world, and finally, what processing should occur given each protocol operation.

With this conceptual framework we go on to assess some existing technical implementations and net protocols that exist for RDF. We primarily consider URIQA[1], RDFNetAPI[2] and Aidministrator[3] as examples of early work in this area. For each of the proposals we evaluate how well they perform against the different criteria of our conceptual framework.

Having surveyed the existing protocols for the world of RDF we turn our attention to Topic Maps. In many other regards Topic Maps have reached the same point of maturity as RDF in all areas except distributed processing. We outline the core differences between RDF distributed processing and that related to topic maps. With this in mind we state the specific requirements that a distributed topic map protocol should satisfy.

With the problem and requirements stated we first present a Topic Map Fragment Proposal. This proposal describes what a fragment is and how they should be generated. The Topic Map Fragment is the core building block of the remote topic map protocol (RTP). One of the core ideas we deal with in this paper is the nature of remote identities of topic map constructs and what default merging or processing can be achieved with no additional operational semantics.

The Remote Topic Map Protocol consists of the definition of a number abstract operations that together make up a protocol for querying, navigating and updating remote topic maps. We utilize the fragment proposal from the previous section and adopt a formal notation for describing the semantics of the operations on the topic maps.

With the fragment proposal and the remote topic map protocol we have a concrete definition of a distributed processing model for topic maps defined in terms of our conceptual framework. We believe that this and subsequent work will define the way in which we can really achieve a semantic web rather than the current island of inaccessible semantic information.