Kelly Boccia Abi Natarajan Konstantin Livitski Senthil Anand Subbanan Meyyappan 1.

Kelly BocciaAbi Natarajan Konstantin LivitskiSenthil Anand SubbananMeyyappan Meyyappan

1

Agenda

Business Requirements Client Overview Business Problem Business Goal  Solution and Scope

Technical Specification System Context  Architecture Overview  Components & Modules Security Model Document indexing Search Explained

Implementation Plan Resource & Costs Development Environment Production Environment Success Criteria 

Prototype Q&A

2

          Multi-National Manufacturing & Sales Corporation  

 Business Growth     - Multiple Applications             - Multiple Repositories           

Business Problem

3

Business Goal 

     Organize Intellectual Capital and Assets 

Accessibility - Connect knowledge workers securely to relevant information

Productivity - Increase productivity and reduce re-work by leveraging knowledge and expertise

Client Overview

4

Solution

Enterprise Knowledge Management Platform

5

System Context

6

Components & Modules

7

Architectural Overview

8

Security Model

• Integrated with existing GLOCO's security infrastructure• Any access requires authentication• To follow a link in search results, user may need additional

authorization for repository access

9

Document indexing

• Document is anything that a search result can point at• Documents are external to the search engine• Documents include text and metadata • Lucene sees each document as a set of named fields

10

How search works

• Lucene sees each document as a set of named fields • A record is created for each document to store some fields

o URL is usually a stored field• The main index is keyed by search term (i.e. inverted)

o Typical text fields are tokenized, filtered, and stemmed into terms o Indexed fields may be discarded after processing o For each term, a list of document IDs is stored to help locate recordso Also stores frequency and proximity

• Search involves retrieval of document IDs by term, and stored fields by the document ID

11

Resource / Cost Plan

 21 weeks total effort 13 member team including GLOCO and Innova INNOVA supports full SDLC with phases

Solution Outline, High Level Design, Detailed Design Build / Test / Deploy and Post Production Support

12

SLATES - Development Environment

Developer workstation to host Virtual Images. Developer workstation to share development

Search Servers Fully configured environment to unit test and

development

13

SLATES - QA / Test and Production

• Sticky load balancer to remember the serving tomcat

 • Each Search server

to hold multiple instances.

 • Shared / Cached

Network storage to share index

 • Similar configuration

for both QA and Production environment

14

Success Criteria and Benchmarks

Most important project success criteria are:  10% time and resource savings on certain R&D activities  75% positive feedback on user surveys  50% of the target user group are actively using the system  5% of available documents have user-defined tags 

15

User 1 Searches for the keyword 'Blood Glucose'

16

User 1 gets back the results with the keyword ‘blood glucose’

17

User 1 adds tag ‘diabetes’ to a result

18

Tag ‘diabetes’ is immediately available for searching

19

User 2 searches for keyword ‘diabetes’

20

User 2 gets back a result for keyword ‘diabetes’

21

User 2 clicks on keyword ‘bp testing’ in the tag cloud

22

User 2 gets more results for keyword ‘bp testing’

23

Thank you!

Innova would like to thank:

Zoya KinstlerJeff Parker

Basem NaseimValar Jayaprakash

Classmates Harvard University Extension School 

24

Questions?

25

Reference Slides

26

27

Index Growth

• Index size is a percentage of the document corpus size• Maintenance trade-off:

o Expensive segment merges - load all segments, write a new oneo Fragmented index is expensive to query - must read all segments

• Lucene index segments are write-once - helps with concurrency• Updates are done as delete - re-add. Updates should be

batchedo Direct tagging is inefficient

28

Scalability

(Source: Mark Miller, "Scaling Lucene and Solr", Lucid Imagination, 2010)

• Query volume is scaled by replication• Index size and indexing load is scaled by sharding

29

Phase 1 - Work Break Down Chart

• 21 weeks total effort• 13 member team including GLOCO and Innova• INNOVA supports full SDLC with phases - Solution Outline,High

Level Design, Detailed Design, Build / Test / Deploy and Post Production Support

30

Use Case - Search and Tag

31

Hardware / Software - Detailed Configuration

32

Interface Specification

33