Efficient result caching mechanisms in search engines

Abstract

The performance of a search engine depends on its components such as crawler, indexer and processor. The query latency, accuracy and recency of the results play crucial role in determining the performance. High performance can be provided with powerful hardware in the data center, but keeping the operational costs restrained is mandatory for search engines for commercial durability. This thesis focuses on techniques to boost the performance of search engines by means of reducing both the number of queries issued to the backend and the cost to process a query stream. This can be accomplished by taking advantage of the temporal locality of the queries. Caching the result for a recently issued query removes the need to reprocess this query when it is issued again by the same or different user. Therefore, deploying query result cache decreases the load on the resources of the search engine which increases the processing power. The main objective of this thesis is to improve search engine performance by enhancing productivity of result cache. This is done by endeavoring to maximize the cache hit rate and minimizing the processing cost by using the per query statistics such as frequency, timestamp and cost. While providing high hit rates and low processing costs improves performance, the freshness of the queries in the cache has to be considered as well for user satisfaction. Therefore, a variety of techniques are examined in this thesis to bound the staleness of cache results without blasting the backend with refresh queries. The offered techniques are demonstrated to be efficient by using real query log data from a commercial search engine.