• information retrieval
• latency
• parallelism
• search engine

In this thesis, after performing an analysis of state-of-the-art techniques for query processing, adopted in both commercial and experimental information retrieval systems, we develop our own implementation. Our implementation is based on a combination of best practices in query processing presented in literature, and aims to reduce the response time experienced by individual queries submitted to the system, without affecting significantly the system throughput. More in detail, we explain the factors that affect the time a query is going to take to be processed and implement a mechanism to exploit these factors to determine the more appropriate degree of parallelism to use to process a given query. Further, we implement a resource manager that handles the simultaneous parallel processing of multiple queries, where each query is in turn potentially divided in multiple tasks. An algorithm for achieving a good work balancing among threads involved in processing the same query have also been implemented.