WHAT HAPPENED TO FAST?FAST was a true Enterprise Search platform, so…
The evolution of FASTFDS ESPFSISFSIAFS4SPSearch inSP2013Search inSP2010Secret sauce(incl. Mars)
End of an era, birth of a New age• FAST now “fully integrated” into SP2013– True, but there’s more!• No longer a “FAST license”– SP2013 contains everything– Enterprise version1997 – 2013
Empoweringthe wholeSharePointexperienceBetter, morepowerfulextensibilityMajor userexperienceoverhaulMain themes of the new Search• User Experience is finally a key componentVastlyimprovedsearch coreFinally asingle searcharchitecture
User Experience is finally key• Revamped user/admin interface• Hover panels, previews• Query rules, result blocks• Result types, display templates• “You’ve seen this result before”• Query Builder• Content Search web part• Etc.
For the first time,Search isn’t defined by thenuts and bolts,but from the User Experienceand high-level tools around it.
Keeping it all togetherServices ProcessesProcess name Descriptionhostcontrollerservice.exe Process controller. Monitors and restarts children.noderunner.exe A search component (except the crawl component)mssearch.exe The crawl component.
Crawl component• Changes from SP2010– Improved performance– Continuous crawl• Improves freshness– Crawl Log• More details• Document removal– Crawl Health Report• Huge improvement!Crawlmssearch.exe
Continuous crawls• Not event-driven indexing• Starts crawl regardless of prior crawl session• Large change sets no longer bad for freshness• Only available for SharePoint content types– Possible to crawl SP 2010 and 2007timeContinuousFull andincrementalDefault 15 min
Crawl health reportsRate Latency FreshnessCPU andmemoryloadContentProcessingactivityEtc.Crawl rate per type Crawl load
Content processing component• Schema mapping– Crawled Managedproperties• Entity extraction– Companies and custom• Advanced Filter Pack is gone– PDFs are out of the box• Extensible through web service• Internally: processing flows– Replaces Python pipelineLink
Index component• Propriety disk-based index• Discrete portions calledpartitions• 1 partition per 10M docs• Each partition contains 1+replicas for fault-toleranceand query volume• 1 replica, 1 server• All servers perform indexing(partially in-memory)
Example: Partitions and replicasSame contentDifferent content
Query processing component• Prepares the queries– Query rules– Result sources– Linguistics/dictionaries– Etc.• Manipulates the results– Display templates– Late security trimming– Etc.• Internally: processing flows– No custom processing as in Content Processing– Still MAJOR improvement
Query rules• For a certain term trigger certain action:– Add/change query terms– Use alternate sorting/relevance– Hybrid search (or other federated results)– Etc.• Replaces search keywords in SP2010• Configure at farm, site collection or site-level
Query builder• Easily builds KQL– CSWP, result sources, query rules, etc.
Query client types• Adjust throttling per client type
Query health reportsTrend OverallLatency inmain flowLatency ineachsubflowIndex times Etc.Latency per processing node in SharePoint flow
Analytics processing component• Analyzes crawled items and search usage• Updates index without re-indexing documents• Result: relevance becomes self-learning– Also: search reports and recommendationsLinkAnalyticsReporting
Search reports• Self-learning relevance aside,never underestimate manual effort!– Query rules, synonyms, boosts, etc.• Automatic reports:– Number of queries– Top queries– Abandoned queries– No-result queries– Query rule usage
Search administration component• Provisions other search components• Talks to Admin database on behalf of:Crawl, Content and Query processingcomponents• In previous FAST products, it was hard to makethe admin component redundant– Not the case in SP2013!– Scale appropriatelyAdmin
Hardware properties• Highlights– In-memory technology– VMs now supported for production– SANs less problematicComponent CPU Memory Disk I/O NetworkCrawl Medium Medium Medium HighContent processing High High MediumIndex High High High MediumQuery processing Low Medium MediumAnalytics processing Medium Medium Medium HighSearch administration Low Low Low Low
Questions?Håll utkik på vår blogg!blog.comperiosearch.comEmail: firstname.lastname@example.orgTwitter: @marcjohaBlog: http://blog.comperiosearch.comLinkedIn: http://www.linkedin.com/in/marcusjohansson