How to tackle performance issues when implementing high traffic multi-language search engine with Solr/Lucene

05/27/2014 - 12:20 to 13:00
Frannz Club
long talk (40 min)

Session abstract: 

This presentation will summarize the experience gained during an amazing journey of our team that implemented, deployed, and monitored the new search platform in production, which replaced a proprietary search engine with the popular open source Apache Solr.

Our company Kelkoo is an e-shopping platform that connects merchants and customers in different countries all over the world.

The core of this platform is the search engine that allows clients and partners to execute full-text queries in order to find the best offers for their search. The queries could be pretty complex: range, filter & function queries, facets etc. They are executed on indexes of more than 15 millions of documents.

We used scalable and feature-rich technologies (Apache Solr/Lucene) to implement the search platform.

We had to deal with exciting problems ranging from which features to implement, how to scale out and up the system, to SOLR and JVM tweaking in order to guarantee fast responses with high traffic on search cluster.

Learn :

- how to explore JVM settings, index merge factors, cache settings, BIOS and RAID configuration

- how to monitor hardware resources : CPU, memory, I/O disks

- how to configure SolrCloud to make best use of hardware (optimize number of queries per seconds and response time)

- what is the impact of SOLR features implementation on performance