On the Response Time of a Solr Search Engine in a Virtualized Environment

This paper describes the results of measured response times of a full-text search service running in a virtualized environment. Using a monitoring tool, we observed the performance of the lulu.com Solr search engine in order to identify a relationship between the CPU and memory allocated to the Solr VM, the arrival rate of queries, and the 95th percentile and the mean of the query execution response time. Based on collected data, we then fitted an analytic expression which relates the 95th percentile and mean of the response time to the CPU allocated to the VM and arrival rate of queries. These expressions can be used for capacity planning of the Solr search cluster at Lulu.