Tarun Jain, ABB Inc, Extreme Faceting using SOLR Case study at ABB Inc © ABB Group 18 March 2018 | Slide 1
About ABB Inc § § World’s largest producer of indutrial robots § Presence in 100+ countries § © ABB Group 18 March 2018 | Slide 2 Global leader in Power & Industrial Automation technologies 2009 revenues USD 33+ billion
About “ABB Products” § § We maintain the master data related to Product classification § We maintain master product attribute data for ~485, 000 products with 21 million+ attributes § Classification tree structure has 45, 000+ nodes and maintained in 31 languages § Product attributes are translated in 10 languages § © ABB Group 18 March 2018 | Slide 3 “ABB Products” is the central repository of all product/catalog information within ABB Connected to 18 different downstream applications within ABB and 5 external applications
About “Product Information Services” § § Started in 2008 § Provides ABB Product catalog search services to downstream applications § ABB. com & ABB Business. On. Line (BOL) are main consuming applications. § © ABB Group 18 March 2018 | Slide 4 Sub-project of ABB Products Several more applications in pipeline to start using search services
Details about Product Information Systems § 6+ million hits per month from abb. com & ABB BOL § 420, 000 items with 20+ million attribute values indexed § 1200+ attribute types § 31 Languages § Running on 2 load-balanced dual-processor quad-core machines with 16 GB of RAM § Software used are § § ASP. Net front-end used to create Web. Control § Backend web-services using ASP. Net & WCF § SOLR 1. 4 using Tomcat with 3 gb RAM § © ABB Group 18 March 2018 | Slide 5 Windows Server 2003 OS Oracle DB
Product Information Services - Features § § Advanced text search services § Browsing services (Navigation through Classification Trees) § Facet search filters § Attribute Group List Resolution to Classification Nodes § One general Web Control to support Navigation, Faceting/List page and Item Detail page § Web Services § © ABB Group 18 March 2018 | Slide 6 Text search services Accurate hit counts everywhere
Challenges § § Faceting on any of the 1200+ attributes § Hit counts needed to be accurate § Support ever growing languages § Same codebase for all 3 major consuming application § Index updates at-least 3 times a day § Average response time less than 500 ms § © ABB Group 18 March 2018 | Slide 7 Search & Classification tree results to be filtered on Country, Customer, Consuming application etc. . And most importantly. . “Everything should be always fast”
Solr vs “Large Commercial Vendor”… Fight !! § SOLR was compared to another major commercial product § Stress test results in Proof of concept… § SOLR 35 req/sec vs 2 req/sec § Average response times 200 ms vs 1 -7 secs § CPU usage 2 -3% vs 100% § § Conclusion. . Performance of SOLR is inversely proportional to the cost § © ABB Group 18 March 2018 | Slide 8 Sadly matchup was not even close (at least for the scenarios we tested for) Winner – SOLR by a KO
Observations § Going from SOLR 1. 3 to 1. 4, faceting performance improved 1. 5 x 2. 0 x § SOLR has no issue with scaling to 1500+ facets § Java based index replication is faster than rsync (atleast on windows) § Tagging filters and excluding them during faceting is cool feature & super useful § While using REST api use chunky instead of chatty calls § Configure Tomcat/Java to minimize GC as much as possible § Do periodic cleanups of SOLR schema to minimize stored values, indexed values, optimize types & fields etc. . Minimize unused bits in the index § Minimize xml data to/from SOLR § Store all lookup data in memory. Cache as much as possible § Minimize database calls © ABB Group 18 March 2018 | Slide 9
DEMO © ABB Group 18 March 2018 | Slide 10
Скачать презентацию Tarun Jain ABB Inc Extreme Faceting using SOLR
cd3b580b7b49a8155d07d400b0aff173.ppt