5/23/2013
Use case for implementing Operational Business Intelligence g
TDWI 2013 Munich TVP - Technology ec o ogy Department epa t e t a and d Atos tos Poland oa d 1
5/23/2013
Topics •
Why OBI is important for business
•
How to prepare business justifications
•
What are operational requirements and standard solutions
•
How to design High Level Architecture
•
What are required components
2
5/23/2013
Presentation case
▶Described solution was delivered by TVP - Technology Department and Atos Poland ▶Solution covers optimization for video content distribution on edge d servers off TVP VOD system t ▶Solution was built in the 2012 H1
3
Telewizja Polska S.A. (TVP S.A.) Polish Public Television Broadcaster
Technology Department
5/23/2013
Media Technologists General, regional, HD and speciality channels: TVP 1, TVP 2, TVP Polonia, TVP Info, Info TVP Sport, Sport TVP Historia, Historia TVP Kultura, TVP HD, TVP Parlament oraz TVP Seriale. New technologies: VOD, live streaming, SmartTV, HbbTV.
4
New European IT champion with global reach
Your Business Technologists
5/23/2013
Powering progress At Atos we strive to create the firm of the future. We believe that bringing together people, technology and b i business iis th the way fforward. d Every day we power sustainable progress for our clients and partners, the wider community and ourselves. It is our unique approach as business technologists that makes this possible. Thierry Breton, CEO Atos
5
5/23/2013
Speaker Data and Information Management Warsaw; Poland
Unlock the hidden value of your data and information Tomasz Bawor DW/BI Solution Architect tomasz bawor@atos net
[email protected] @tbawor
6
Why OBI is Wh i iimportant t t for f business
7
5/23/2013
Operational Business Intelligence d fi i i definitions
5/23/2013
▶ A set of services, applications and technologies for monitoring, reporting on, analyzing and managing the business performance of an organization’s organization s daily business operations “Embedded BI” By Judy Davis and Colin White, www. B-EYE-Network.com Research 2008 paper
▶ Right time Business Intelligence “Right-Time Business Intelligence: Optimizing the Business Decision Cycle” by Judy Davis, www. B-EYE-Network.com Research paper (2006)
▶ Adaptive Business Intelligence The phrase was coined in 1999 by SolveIT Software Chairman and co-founder Dr. Zbigniew Michalewicz
▶ Decision Management Systems “Decision Management Systems: A Practical Guide to Using Business Rules and Predictive Analytics” By James Taylor Published: Sep 30, 2011
8
What do we mean by Operational B i Business IIntelligence lli ▶ Operational system and Business Intelligence combined to drive better performance for the business process ▶ Operational and BI technology stack which should be used in most optimized way
9
5/23/2013
What is the added value
PREPARE DATA
ACT
5/23/2013
MONITOR
▶ Performance improvement is the concept of measuring the output of a particular process or procedure, then modifying the process or procedure to increase the output, increase efficiency, or increase the effectiveness of the process or procedure. http://en.wikipedia.org/wiki/Performance_improvement
▶ Basically we would like to decrease resources and/or increase results. Other definitions based on simmilar pillars: BPM, Kaizen, Six Sigma
10
When OBI is most suitable
5/23/2013
▶ When there is impact on performance by: ▶ Integrate additional data ▶ Store historical data ▶ Use Advanced Analytics based on sophisticated p calculations ▶ Use High Performance DW/BI or Big Data technology
11
TVP business context
5/23/2013
▶ Video on Demand (VOD) system ▶ Video content could be broadcasted through: laptop, notebook or PC with www; Tablet, SmartPhone, SmartTV with dedicated application ▶ About 50 thousands different content per month ▶ About 6 millions requests for content per day
12
TVP business requirement
5/23/2013
Improve p Quality y of Service ((QoS)) for broadcast popular video content without stop video playback 13
TVP technology context
5/23/2013
▶ VOD system contain few servers with two layers: ▶ Fast based on SSD drives ▶ Slow based on SATA which is 20 times bigger than SSD layer
▶ Cache management (SWAP) which contains: ▶ SWAP IN – video content is loaded from SATA into SSD ▶ SWAP OUT – last played video content is removed from SSD each time when there is no space left
▶ High volume of low popular content were mixed with high popular on the same servers
14
TVP technology requirements
5/23/2013
▶Decrease number of cycles for SWAP IN and SWAP OUT for popular content ▶Split content based on the popularity through servers
15
5/23/2013
TVP solution ▶ Prepare p and store
▶ Additional data about content with information about g y hierarchy y category ▶ Calculations for content popularity in different dimensions
▶ Learn and forecast content popularity ▶ Act by delivery of whole list how to distribute content ▶ Monitor content popularity between distributions and modify distribution list if necessary
16
How tto prepare b H business i j justifications
17
5/23/2013
Start from business process
5/23/2013
▶ Each management level has its own Set Goals and Monitor
Strategic
point of view. For operational level we should proceed different than on above above. ▶ Strategic -Gather information and options
Build Plans and Notify
Tactical
▶ Tactical - Integrate information according to strategic goals
Operationalize and Act
▶ Operational - Design business process Operational
according g to tactical p plan
18
Model business process
5/23/2013
▶OBI could act on step level of business processes ▶OBI is complementary to legacy systems ▶OBI acts on hard to change decision making components of legacy systems with more manageable, more agile components
19
Characteristics of suitable d i i decisions
5/23/2013
▶ Repeatable ▶ The decision is made at defined times; the same information is used each time; defined set of actions; consistent measurement of success
▶ Nontrivial ▶ Policies and regulations; domain knowledge and p analysis y is required; q large g amounts of expertise; data; large numbers of actions; trade-offs must be made; continued updates
▶ Measurable business impact ▶ Candidates for automation 20
5/23/2013
Time to M T Market
Prioritize business processes to change h
Build roadmap starting from small changes with big g impact p 21
Solution evaluation
5/23/2013
▶Find all the technical alternatives ▶Evaluate these alternatives ▶Select the best alternative for implementation ▶OBI should be more agile, manageable, reliable and simple ▶Start hypothesis testing as soon as possible
22
TVP technical alternative evaluation l i
5/23/2013
▶ Cache management algorithms: ▶ Least Recently Used (LRU) – SWAP OUT for last used content ▶ Least Frequently Used (LFU) – SWAP OUT for least popular ▶ Adaptive Replacement Cache – combination of LRU and LFU (Patented by IBM in 2006)
▶ Content aware systems: ▶ Content Delivery Network – geographically extensive network for content distribution ▶ Transparent/Edge Caching – keeping content close to the target audience ▶ Adaptive Content Delivery – isolation of content with unpredictable characteristics, which results in increased traffic on the network
23
TVP content hypothesis testing D Does content iinsight i h add dd any value? l ?
5/23/2013
Drop Page Fields Here MATERIAL ID MATERIAL_ID
3000
4309239
Sum of REAL_REQ_NUMBER
7811575 7821552 7832365 7832416
2500
7832441 7863829 7863915
2000
7864003 7864056 7864079
1500
7864227 7892292 7892348
1000
7892369 7903760 7903804
500
7913348 7913404 7923528 2012-09-24
2012-09-23
2012-09-22
2012-09-21
2012-09-20
2012-09-19
2012-09-18
2012-09-17
2012-09-16
2012-09-15
2012-09-14
2012-09-13
2012-09-12
2012-09-11
2012-09-10
2012-09-09
2012-09-08
2012-09-07
2012-09-06
2012-09-05
2012-09-04
2012-09-03
2012-09-02
2012-09-01
2012-08-31
2012-08-30
2012-08-29
2012-08-28
0
7923593 7923614 7929935 7929984 7930008
DATE
7944502
Characteristics of the information program 24
2012-07-14 2012-07-27 2012-07-28 2012-07-29 2012-07-30 2012-07-31 2012-08-01 2012-08-02 2012-08-03 2012-08-04 2012-08-05 2012-08-06 2012-08-07 2012-08-08 2012-08-09 2012-08-10 2012-08-11 2012-08-12 2012-08-13 2012-08-14 2012-08-15 2012-08-16 2012-08-17 2012-08-18 2012-08-19 2012-08-20 2012-08-21 2012-08-22 2012-08-23 2012-08-24 2012-08-25 2012-08-26 2012-08-27 2012-08-28 2012-08-29 2012-08-30 2012-08-31 2012-09-01 2012-09-02 2012-09-03 2012-09-04 2012-09-05 2012-09-06 2012-09-07 2012-09-08 2012-09-09 2012-09-10 2012-09-11 2012-09-12 2012-09-13 2012-09-14 2012-09-15 2012-09-16 2012-09-17 2012-09-18 2012-09-19 2012-09-20 2012-09-21 2012-09-22 2012-09-23 2012-09-24 (blank)
TVP content hypothesis testing D Does content iinsight i h add dd any value? l ? 2500
DATE
Characteristics of the popular TV series 25
5/23/2013
Drop Page Fields Here MATERIAL ID MATERIAL_ID
Sum of REAL_REQ_NUMBER 3994796
3994809
3996934
3996954
2000 3996956
4071827
4071831
4080896
1500 4080900
4119634
4119646
1000 4165790
4165792
4198836
4198863
500 4198866
4239309
4362349
0 4362350
4442927
4444267
4444271
4480765
4480777
4519317
4525473
2012-07-14 2012-07-27 2012-07-28 2012-07-29 2012-07-30 2012-07-31 2012-08-01 2012-08-02 2012-08-03 2012-08-04 2012-08-05 2012-08-06 2012-08-07 2012-08-08 2012-08-09 2012-08-10 2012-08-11 2012-08-12 2012-08-13 2012-08-14 2012-08-15 2012-08-16 2012-08-17 2012-08-18 2012-08-19 2012-08-20 2012-08-21 2012-08-22 2012-08-23 2012-08-24 2012-08-25 2012-08-26 2012-08-27 2012-08-28 2012-08-29 2012-08-30 2012-08-31 2012-09-01 2012-09-02 2012-09-03 2012-09-04 2012-09-05 2012-09-06 2012-09-07 2012-09-08 2012-09-09 2012-09-10 2012-09-11 2012-09-12 2012-09-13 2012-09-14 2012-09-15 2012-09-16 2012-09-17 2012-09-18 2012-09-19 2012-09-20 2012-09-21 2012-09-22 2012-09-23 2012-09-24 (blank)
TVP content hypothesis testing D Does content iinsight i h add dd any value? l ? 500
DATE
26
5/23/2013
Drop Page Fields Here MATERIAL ID MATERIAL_ID
Sum of REAL_REQ_NUMBER 5962293
5980705
450 5999150
5999311
400 5999437
6158442
350 6158458
6158514
300 6158529
6163466
250 6163511
200 6163595
6163758
150 6163892
6291308
100 6294064
50 6294083
6294088
0 6294106
6294136
6294996
6295467
6314322
6318215
6445099
6445116
Characteristics of the TV series with long history
Build equation to compare Champion Challenger Testing
Production Proces
Output
5/23/2013
▶Champion Challenger testing when there is possible to test in
Input Background g Proces
Output p
A/B Testing
background on the whole production data
Production Proces A
Output
Production Proces B
Output p
Input
▶A/B testing when there is possible to test on the splitted production data
Statistics Testing
▶Statistics testing when above are
Production Proces Input
Output Gather Statistics
impossible 27
TVP solution assessment
5/23/2013
Chosen statistics testing g based on forecast accuracy y due to no possibility to test old and new solution in the same time
28
5/23/2013
What are operational requirements and standard solutions
29
Operational requirements
5/23/2013
▶ Prepare Data ▶ Real Time – the same state as on the operational system; Near Real Time – state with minimal latency
▶ Store Data ▶ High Performance Databases
▶ Learn ▶ High Performance Advanced Analytics
▶ Monitor ▶ Full automatization; Full accountability
▶ Act ▶ High Availability; High number of requests 30
Standard solutions
5/23/2013
▶ Prepare Data ▶ Stream/event processing; Minimal batches
▶ Store Data ▶ In memory databases
▶ Learn ▶ In database Advanced Analytics
▶ Monitor M it ▶ Messages; SLA management
▶ Act ▶ Business Rules Management; Failover; Load balancing; SOA
31
TVP operational challenges
5/23/2013
▶Variety of applications resulting in different representations in the logs ▶Internal I t l algorithms l ith resulting lti iin complex system behavior and logs representation ▶Variety of content with different metadata 32
How tto design H d i Hi High hL Levell Architecture
33
5/23/2013
5/23/2013
Single EDW Set Goals and Monitor
Self Service Dashboards
Build Plans and Notify
Self Service Ad Hoc
Operationalize and Act
Self Servise Advanced Analytics
Operational Business Intelligence
Full Service Predefined Reporting
Full Service Advanced Analytics
Business Rules Management g
Data Marts
Enterprise Data Warehouse
Big Data
ODS
Cost of creating such a low-latency BI environment may be more than the actual benefit the company receives 34
5/23/2013
Federated EDW
E h componentt used Each d
Replication n
only for its purpose with data replication p between them
35
5/23/2013
TVP before OBI ▶ Play video content:
▶ User requests video content through www ▶ DISPATCHER directs to proper server ▶ Content if available in CACHE (SSD) than broadcasted ▶ Content if not available in SSD than loaded from SATA into SSD
▶ Algorithms: Al ith ▶ DISPATCHER based on hash algorithm to distribute content ▶ SSD and SATA management based on internal application server algorithm
36
5/23/2013
TVP with OBI
▶DISPATCHER is based on OBI list • OBI list directs each video content on proper server • Additional Additi l metadata t d t about content used for calculations and learning
37
5/23/2013
What are required components
38
5/23/2013
OBI components
Each of components should be used only if it is necessary therefore whole development process will be agile and fast 39
TVP used components
5/23/2013
Data Store – stores dedicated calculations; Advanced Analytics – once per day prepares list with forecast; Business Rules Management – through day updates list after corrections from ongoing broadcast; Notification – sends alert if there is an issue; Automatization – whole process without ith t h human iinteraction t ti 40
5/23/2013
Conclusions
41
5/23/2013
Further reading
42
5/23/2013
Questions Atos IT Services sp. z o.o. ul. Postępu 18 (budynek Neptun) 02-676 02 676 Warszawa tel. +48 22 444 6500 fax. +48 22 444 6501
Tomasz Bawor Solution Architect
[email protected]
For p public q questions and comments after session please use twitter @tbawor 43
5/23/2013
Thank you y Atos, the Atos logo, Atos Consulting & Technology Services, Atos Worldline, Atos Sphere, Atos Cloud, Atos Healthcare (in the UK) and Atos Worldgrid are registered trademarks of Atos SA. September 2011 © 2011 Atos. Confidential information owned by Atos, to be used by the recipient only. This document, or any part of it, may not be reproduced, copied, circulated and/or distributed nor q quoted without p prior written approval from Atos.
44