Distributed Selective Dissemination of Information

To help users cope with information overload, Selective Dissemination of Information (SDI) will increasingly become an important tool in wide area information systems. In an SDI service, users post their long term queries, called pro les, at some SDI servers and continuously receive new, ltered documents. To scale up with the volume of information and the size of user population, we need a distributed SDI service with multiple servers. In this paper we rst address the key problem of how to replicate and distribute pro les and documents among SDI servers. We draw a parallel between distributed SDI and the well-studied replica control problem, adapt quorum-based protocols for use in distributed SDI, and compare the performances of the di erent protocols. Next we address another important problem, that of e cient document delivery mechanisms. We present and evaluate a practical scheme, called pro le grouping, which exploits the geographical locality of users to cut down network tra c generated by document delivery. Finally, we carry out a sensitivity analysis to determine the parameters that have critical impact on performance, and investigate strategies to cope with the scaling up of those parameters.