• Accelerated migration from legacy proprietary search solution to Solr in 30 days
  • Optimized search strategy with help from Lucid Imagination by refining data and document structures, reducing search field count by 200x
  • Fine-grained control of search relevance drives revenue increase by keeping customers on-site

Overview

An automotive search and information services company has over a dozen Web properties, each targeted at a different of segment of cars and drivers. Sites operated by the company variously help consumers find the cars they need, compare prices and configurations, easily find industry news and information, interact with each other via social technologies and more. Search is a mission-critical component of the company’s advertising-driven business model. Controlling the quality of search on site ensures that people stay on site longer, and more time on site translates directly into increased revenue directly.

Solr is used in several of the company’s web properties, each of which has its own unique search and data needs. Another of the sites uses a proprietary search server, which was targeted for migration to Solr. The company engaged Lucid Imagination to provide expert recommendations on the migration, as well as on optimizing its Solr search infrastructure, performance, relevancy, and business continuity.

Challenges

In the automotive marketplace, information such as models, options, consumers, and suppliers changes every year, and the company’s maintains a sophisticated system to keep track of and present requested information. The company’s content management system presents articles, photos, and videos. Articles are stored in XML while metadata, such as video transcripts, is also stored in XML. The XML is loaded into JavaBeans objects, which are then indexed into Solr using SolrJ.

Because users often find the company’s sites via external internet properties, such as Yahoo and Google, keyword search is vital to keeping them on the site. Ensuring that success search refinements continue to provide relevant results, given the ad-driven model, is strongly tied to revenue. Search on the site is revenue-critical, and hence, mission-critical.

Solutions

Lucid Imagination was engaged to assess several of the company’s Solr implementations. While the company’s search development group had good experience with Solr, it wanted to implement the changes very quickly, and Lucid’s analysis provided assurance that an accelerated schedule could be achieved.

One of the Solr implementations, on one web property, was targeting replacement of a legacy proprietary search system. Lucid reviewed the plan to ensure it would achieve its technical requirements and business objectives. The company wanted to highlight different content types (video, photos, and/or text) in search results, and the Lucid team analyzed the search application they had built to ensure that results would include this content appropriately. To improve stability, Lucid evaluated and recommended an approach that used sharding and replication, along with guidelines for how implement each of these within the scope of performance requirements.

Another company property, already used Solr for search, but turned out to require redesigning of the schema, according to Lucid’s analysis. This change dramatically reduced the number of search fields — by a factor more than 200x, with a significant impact on search footprint and throughput. Assistance on localSolr helped add a dimension of geo-specific relevance to search results.

After reviewing the search infrastructure, the team also made further recommendations: where to use multiple Solr cores; where single instances would be optimal; and key factors to consider when determining how to balance performance and hardware. The company’s developers had written some custom code for its current Solr 1.3 deployment; Lucid verified this custom code would work as intended on the new Solr 1.4 release, along with identifying other benefits of working with the new release.

Based on Lucid’s recommendations, the web property using the legacy proprietary commercial search tool was successfully migrated to Solr by the customer within 30 days of the engagement. With Lucid’s expertise, the company also optimized its search deployments across different aspects of its web properties, meeting performance and reliability goals in the face of the unique content mix and interdependencies of the automotive

Tags: localSolr, SolarJ, car, automotive, migration, Solr 1.4, replication, sharding, revenue, relevance.