Serval is an experimental testbed that we are building to measure the performance and scalability of our mobile-agent system. The testbed consists of two components: a hardware platform, and a software application.
The hardware platform is a cluster of 25 Linux workstations, connected by two Fast Ethernet networks. For Serval, that cluster acts as a parallel server. In another lab, we have a collection of 15 Linux laptops, connected by Ethernet to the server cluster. We use the laptops to run simulated clients, and the server cluster to run our information-retrieval server.
The application is an example of a typical information-retrieval engine. We collected USENET news articles for several weeks, accumulating several gigabytes of documents. We indexed the documents by the words they contain, and enter the words in a MySQL database. The database is spread across all nodes of the server cluster. We then provide two different access methods to the database:
The basic system is complete, as is a simulation framework that allows us to run the same set of queries against either the RPC or the Agent implementation, and compare them carefully. We are in the final stage of integration and testing, and should have initial results in August.
There are no papers specifically about Serval. These other papers address the information-retrieval aspects of mobile agents.
Mobile Agents for Distributed Information Retrieval: Brian Brewington and Robert Gray and Katsuhiro Moizumi and David Kotz and George Cybenko and Daniela Rus. In Matthias Klusch, editor, Intelligent Information Agents, chapter 15, Springer-Verlag, 1999.
Transportable Information Agents: Daniela Rus, Robert Gray and David Kotz. In Michael Huhns and Munindar Singh, editors, Readings in Agents, chapter 3.3, Morgan Kaufmann Publishers, October, 1997.
Relevant projects, inside D'Agents: