Solr Implementation - TMP Worldwide Advertising

TMP Worldwide Advertising is the largest independent firm focusing on recruitment advertising and communications. TMP offers a depth of resources and services including digital expertise, brand development as well as traditional creative and targeted media planning. TMP deploys a search-engine-optimized index of searchable jobs and has enjoyed strong growth of its service offering to its many client companies. However, the proprietary search solution employed for the public-facing site began to show some strain, occasionally causing serious problems and partial outages of service. As such, TMP decided that a more robust and powerful search solution was necessary. TMP processes XML feeds of job data provided by their client companies. This data is extracted and loaded into a Microsoft SQL Server database and serves as the primary content for job data on the public-facing web site. During the processing of this data, the existing proprietary search data was generated and inserted into the database. At the time of this project the total number of jobs available averaged around 20,000.

The project tasks were essentially organized around four themes: process client data quickly with minimal impact to production resources, improve the quality of search results, provide a highly-available and scalable system, and take ownership of an easy-to-grow and maintainable solution by staff.

Innovent delivered the project on time and on budget. Search results were reduced from 2-6 seconds on average to less than 250ms on average. This led directly to the search results having a highly-noticeable difference in speed, and being “lightning fast”. The complex and flexible field boosting capability and natural language processing features of Solr provided excellent tools to swiftly address interesting search scenarios for its diverse client base. Most importantly, the strain on the database to process and provide search results was eliminated, moving search processing time down from 3 hours to less than 20 minutes and eliminating the highly-contentious machine processing necessary to provide search.

Click here to read the full story (pdf).