Vertical search engines focus on specific slices of content, such as the Webof a single country or the document collection of a large corporation. Despitethis, like general open web search engines, they are expensive to maintain,expensive to operate, and hard to design. Because of this, predicting theresponse time of a vertical search engine is usually done empirically throughexperimentation, requiring a costly setup. An alternative is to develop a modelof the search engine for predicting performance. However, this alternative isof interest only if its predictions are accurate. In this paper we propose amethodology for analyzing the performance of vertical search engines. Applyingthe proposed methodology, we present a capacity planning model based on aqueueing network for search engines with a scale typically suitable for theneeds of large corporations. The model is simple and yet reasonably accurateand, in contrast to previous work, considers the imbalance in query servicetimes among homogeneous index servers. We discuss how we tune up the model andhow we apply it to predict the impact on the query response time whenparameters such as CPU and disk capacities are changed. This allows a managerof a vertical search engine to determine a priori whether a new configurationof the system might keep the query response under specified performanceconstraints.
展开▼
