Note: This page is historical.

Current pages about Yenta are here. Please look at those pages first.

Yenta is still under active development, but this particular page is not. If you're interested in current research papers about Yenta, or obtaining a copy of Yenta, please start here instead.

This page is one of many that were written in late 1994 and early 1995, and are being preserved here for historical purposes. If you're viewing this page, you probably found it via an old link or are interested in the history of how Yenta came to be. These pages have not been actively maintained since 1995, so you'll find all sorts of older descriptions which may not match the current system, citations to old papers and old results, and so forth.

Bootstrapping an agent into the community

When a Yenta agent first comes up, it must somehow find other agents with which to communicate. How can it do so? There are several heuristics:

If the agent has been used before, check the stored list of other agents with which it was communicating.
Ask the user who else might be running a similar agent, and then try to contact those users on the indicated machines.
Look in various files owned by the user (for instance, the .mailrc file, which is used by several UNIX mail programs to store shorthand names for other email addresses the user corresponds with frequently) for clues, then try to contact those locations.
Attempt random connections to other machines in the domain, until one of them reports that it, too, is running an agent.
Consult a central registry of starting points, maintained in the same place as the code distribution of the agent.

None of these heuristics are perfect. Certainly, if this is the first time that the user has run the agent, and knows of nobody else who is running one, few options are attractive. While the option of asking the distribution server for clues seems attractive, such an option should be a last resort, since otherwise the central server may become overloaded. Once the agent has successfully contact another agent, even once, its chances improve dramatically, of course. A first contact yields both the contactee other agent, and whatever other agents it may know about, which yields a set big enough to start querying on startup. Even if some of those agents are down at the moment, it is likely that someone will be up, which is all that is required to bootstrap. It is still possible, of course, for an isolated group of agents, which know about each other but not about the rest, to exist inadvertently. However, unless such a group is deliberately isolated behind a firewall, it seems likely that any given island of agents will eventually come into contact with another, in the same way that many disparate human social circles often have a member in common.

Lenny Foner

Last modified: Wed Dec 7 20:51:36 1994