P2P Computing with Web Arrays

Is the Web really ubiquitous yet? Its penetration, reach, and connection with the physical world leave a lot to be desired. Interoperability and free access are not at the levels that would ensure the explosion of usage we can call ubiquity. Sorry!

We present an experimental computing model and a platform, grounded in the best principles of the Web – simplicity and decentralization, that might help in taking us there sooner. Throughout, we stick to widely accepted standards and emerging best practices for the Web.

In the center of the platform is the Web array – a richly annotated ordered collection of (i) URLs of Web resources, and (ii) literal content of UNICODE text or arbitrary well-formed XML. Initially, they were meant to contextually encapsulate, objectify, and universally identify Web 2.0 user data. Now they serve as anything from better bookmarks to batch jobs. The benefit is protocol homogeneity and easy scalability.

Web arrays are represented as Atom feeds. Atom’s own ubiquity is a great basis for a computational platform. Atom allows rich metadata in the form of tagging and annotation. <category> tags can be used for flat textual tagging but also to reference ontologies. Ontologies can also be defined via Web arrays! This way we tie tagging folksonomies with higher ontologies and use the best of both worlds – spontaneous self-formation and higher semantic interoperability.

As fully-XML documents, Web arrays are processed exclusively by XML pipelines (in particular A. Milowski’s SmallX, one of the functional prototypes of the emerging XProc specification). A protocol which builds on Atom and is exposed by REST Web services ensures continual simplicity and scalability. We join those who suggest that REST, much more then the SOAP stack, can carry the Web to ubiquity.

By embedding instructions in the literals a la H. Halpin’s “One Document to Bind Them” we can use Web arrays as batch jobs. Knowing how slow decent Atom feed readers have been in appearing, we choose to do the processing on the server by adding pipeline steps. We then distribute the Web array service onto multiple nodes. Each node carries out the basic Web-array-lifecycle functionality but also specializes in additional processing. We layer a P2P component (JXTA and later Jabber, which was chosen by Google as a carrier) onto the core service on each node. SSO and Web-array aggregation is performed in the P2P layer and make the network appear as a unified application to the user. This is the beginning of a computing platform based on Web arrays.

Web arrays are lightweight objects suitable to be manipulated on a variety of devices. In particular, we are very interested in geo-tagging by agents on mobile devices (which really have become ubiquitous) and using our platform to solve problems in information fusion in multi-agent systems.

Web arrays (Atom feeds) are stored in native XML storage on the peers. We plan to release the single-peer package as open-source when it takes shape, along with an AJAX API. We plan to write applications that will allow users to have a local copy of their Web 2.0 data. This will be our contribution to the open data movement, which has become a necessity in the wake of the Web 2.0 burst.