Information retrieval systems are challenged to handle larger and larger datasets as time passes. Using parallel machines, with each machine holding a part of the inverted index is a solution to decrease the query response times for large scale databases. Central Broker approach is a well known parallel solution using master-slave scheme. A disadvantage of this approach is central broker tries to handle so much work that it can become a bottleneck itself. We propose two modifications of the original central broker schema, namely Distributed Central Broker and Pipelined, which try to decrease the workload of the central broker, by letting the slave nodes do the merging of partial answer sets. We also propose batch query processing, which introduces the problem of query scheduling.

DATE: 5 November, 2007, Monday@ 15:40
PLACE: EA 409