Peer-to-Peer Webstructor

Peer-to-Peer Protocol in Webstructor

Interestingly, there is almost no any specific "protocol" enabling Webstructor servers/applications to talk one to another.

To make this issue clear, let us consider that any protocol has two components - content delivery (which is purpose of the protocol) and auxiliary services (like handshaking, flow control, authentication etc.).

For the latter, Webstructor relies on existent low-level transport protocols, depending on kind of client/server or peer/peer channel. For the access of Webstructor Servlet from Webstructor Applet or Webstructor Application, we use HTTP. For the interaction between Webstructor Applications, we use plain TCP/IP sockets. When authentication is needed, HTTP may be replaced with HTTPS while TCP/IP may be replaced with SSL.

For the content delivery across Servlets, Applets and Applications, Webstructor uses plain ORL statements. This is similar to human-to human interaction where natural language is used. When one Webstructor peer wants to tell something to another, it just sends respective ORL statements over respective transport protocol.

Here is some policy specifying the way ORL statements are used when maintaining conversation between Webstructor peers.

Each time when two Webstructors establish physical connection for the first time, one which behaves as "requestor of connection" sends to "acceptor of connection" ORL statement describing its place in the net (with optional authentication information).

Each time when Webstructors peer accepts physical connection, it waits for "connection request" in ORL (described above) and performs optional authentication. Then, it sends to other peer complete set of ORL statements expressing current content of its knowledge base.

Each time changes in the knowledge base of one of peers are made, respective one sends to all of known peers set of ORL statements incrementally describing change in knowledge base content.

Obviously, approach described above doesn't scale well for the case with enormous amount of data in knowledge base. From one hand, this may be solved adding "interrogative" component to ORL syntax (which currently has only "declarative" one). Then, each of the peers could send to another one an ORL request intended to retrieve only small subset of the knowledge base.

From the other hand, if we may assume that global shared knowledge may consist of many knowledge bases (maintained by peers) shared to small extent (due to limited data retrieved by rare and specific queries) only. This leads to complex problem of global inconsistency of shared knowledge. That is, if we want global knowledge to be consistent, we need to maximize identity of shared part of content across peers. If so, we need to assure that even if some of peers has been offline for a while, it receives complete set of recent changes when it gets online again.

The problem of efficient and consistent distribution of knowledge specific to every peer in the knowledge sharing network without of having entire set of knowledge duplicated by every peer is tough one. Some problems are discussed on page dedicated to knowledge distribution and some other are in current study.

Contents