How did it go? The first large enterprise search project in Europe using SharePoint 2013
1. How did it go?
The first large enterprise search project
in Europe using SharePoint 2013
Smarta Sök :: Stockholm :: January 29th 2014
Petter Skodvin-Hvammen, Solutions Architect
2. Who am I
Petter Skodvin-Hvammen
•
•
•
•
Senior Consultant, AD-Gruppen
Advisor, Architect, Developer
15 years in consulting
7 years with search (FAST + SharePoint)
Search clients
Live in Tønsberg, Norway
Oldest city in Scandinavia
visittonsberg.com/en
Nokas (no), Aker Solutions (no), L’Oréal (fr), Gyldendal
(no), Media Saturn (de), Chalmers (se), Atlas Copco
(be), Marktplaats (nl), Metro (de), Eniro (no), Genpact
(in),
News International (uk), Aller Group (dk), …
3. Background
• Global oil services
company
• 20,000 employees
• 85 locations in 30
countries
• Grow revenue 15 %
• Grow head count with
13,000 in 4 years
• Company acquisitions
• Global knowledge sharing
• Senior engineers retiring
• Employee onboarding
• Employee turnover
4. Knowledge Initiative
• Knowledge mgmt
organization
• Culture for sharing
• IT investments for supporting
discovery and sharing
•
•
•
•
•
Enterprise Search
People profile sites
Communities of practice
Knowledge sharing portal
Tools for validated information
• Avanade
• Project Manager
• AD-Gruppen:
• Solutions Architect
• Comperio:
• Search Consultants
• Designers
• Developers
5. Key pains
• It’s hard to find own documents
• Only closest colleagues are asked for help
• People don’t search for answers, but others who may know it
• Sharing across projects, business areas and locations are
limited
• It’s hard to find and reuse knowledge assets from the past
• Core systems and applications are not consolidated
60 core systems exist, and more than 2,000 applications are in
use
6. Content Sources
• SharePoint 2007 Intranet
Sites
• SharePoint 2007 Team Sites
• SharePoint 2013
Communities
• SharePoint 2013 My Sites
• SharePoint 2013 People
Profiles
• Windows Server File Shares
• SQL Server Database
• 700,000 items
• 7 M items | 1,000 sites | 2.5
TB
• 20 sites (new)
• 25,000 sites (new)
• 25,000 profiles (new)
• > 100 M items | 3,000 shares
• > 5,000 items
• > 4,000 items
7. Search Strategy
• Create an internal «Google»
• «One search to find it all»
• Find answers by
•
•
•
•
•
Finding people
Finding more than you expect
Finding people by documents
Finding documents by people
Finding people by projects
• Users where looking for
•
•
•
•
•
People
Projects
Products
Documents
Communities
8. Extended Search
Box
User Experience
Verticals from
Information Model
Topic Pages
Result Groups and
Display Templates
Custom Refiners
People Hits in
Search Box
http://intranet/search/pages/results/documentname.docx
http://intranet/search/pages/results/documentname.docx
Custom Feedback
http://intranet/search/pages/results/documentname.docx
12. Server Infrastructure
• 14 virtual machines
• 8 CPU cores (112)
• 16 GB memory (224 GB)
• Increased to 32 GB after go-live
• 150 GB system disk (2.1 TB)
• 450 GB data disk (6.3 TB)
• 200 IOPS (x 8 = 1,600 IOPS)
• 4 physical hosts (VMware
ESX)
• 2 datacenters
• UAT, QA, DEV environments
• Server Software
•
•
•
•
SharePoint Server 2013
Windows Server 2008 R2
SQL Server 2012 Enterprise
Netscaler Load Balancer
• Development Software
• Visual Studio Premium 2012
• Team Foundation Server
13. Infrastructure Tuning
Capacity test findings
Resource monitoring revealed
• Crawl rate declined 1% per
million items indexed
• Query latency increased
exponentially from 12 million
items per partition
• Disk usage significant lower
than expected
• 850GB -> 450GB = 5.6 TB
• Vmware:
• Don’t allocate more CPUs than
your virtual machines need
• Don’t dedicate more memory
than available when failover
occurs
• Optimizations are planned
14. Performance Considerations
• Add enough hardware
resources (CPU, Memory, IO)
• Give SQL Server enough
memory
• Exclude index location from
virus scanning
• Know your virtual
environment
• Use separate servers for
distributed caching
• Design for performance and
go easy on the query rules
15. Improving Relevancy
• Relevancy workshops with
knowledge managers
• Analyzed search reports
• Top queries
• Abandonned queries
• Zero result queries
• Testing and tuning by
business
•
•
•
•
Defined test cases
Managed promoted results
Managed synonyms
Managed query suggestions
• Created custom ranking
models
•
•
•
•
Increase weight for hits in title
Add weight for freshness
Improve weight for proximity
Boost intranet content
• Custom app for rank model
testing by business users
• Only admin has access
• Testers need to select model
for comparing results
16. Current Status
• 25% of index capacity in use
• Performance can be
improved
• Infrastructure to be optimized
• Relevancy can be improved
• Governance established but
not adopted by business
• Many users still don’t like to
see people when searching
for documents, and vice versa
• Unstable solution
• Information architecture
• People still don’t know how to
search effectively
• People don’t think findability
when creating content
• Important content sources not
added due to high complexity
and costs (SAP and ProArc)
17. How did it go?
• 12,000 searches per day
• Top 3: Best SharePoint Search Solution
-- European SharePoint Community Awards
2013
• Microsoft SharePoint Case Study