Tuning integrated dissemination-based information systems

Abstract

In an integrated dissemination-based information system (DBIS), data are disseminated to clients in two modes. In the first mode, data publication mode, frequently accessed data are periodically pushed (published) to clients, who only need to listen to the communication channel and filter the incoming data stream for the desired objects. The second mode, demand-driven mode, employs a request-response mechanism for clients to explicitly request for data that are not published. For an integrated DBIS to be effective, the two modes of information dissemination must be carefully employed. For example, publishing too many data objects can lead to excessively long access time for the publication mode even when the workload (i.e., number of clients) is low. On the other hand, publishing very few data objects can lead to excessive waiting for clients using the demand-driven mode when the communication channel becomes a bottleneck. In this paper, we identify the kinds of bottleneck that may arise in an integrated DBIS, and propose a novel algorithm to tune such a system for “optimal” performance (in terms of average access time). We also study the effect of different feedback mechanisms that can help the server to adapt to changing access patterns. We evaluated the various proposed schemes based on a simulation model, and report our findings in this paper.