Upload
lyliem
View
223
Download
1
Embed Size (px)
Citation preview
The Benefit of Leveraging System z with IBM Cognos 8with IBM Cognos 8
AgendaAgenda
Sh t i t d ti N i• Short introduction Numius• Cognos in Information Management from the
partner perspective• A short functional demonstration of Cognos on g
System z• Results of a client PoC for Cognos on System zResults of a client PoC for Cognos on System z
in a joint effort between IBM SG Bel, IBM Montpellier and NumiusMontpellier and Numius
Numius team
Leading provider of Performance Management Solutions and Services in Benelux.
30 hi hl kill d d f i l lt t30+ highly skilled and professional consultants.
Full range services, from Vision Creation, Business Analysis to ImplementationBusiness Analysis to Implementation, Architecture, Operations and Outsourcing.
Multi-industry, multi-function experience.
Focus on long-term partnerships.
Trusted advisor roleTrusted advisor role.
Numius activitiesNumius activities
© Numius nv© Numius nv
Numius-IBM relationshipp
• 2009 Most Distinguished Achievement Award Southwest IOT• 2009 Belux Partner of the Year• 2008 Only Platinum Partner in BeLuxy• 2007 Platinum Reseller• 2007 Partner of the Year• 2007 Reseller of the Year• 2006 Partner of the Year• 2005 Enterprise Planning Partner Award• 2004 Business Intelligence Partner Award2004 Business Intelligence Partner Award
IBM Advanced Business Partner
Cognos in Information Managementg g
The Performance Management Wheel –Performance Management is Every Employee’s Concern
How are we doing? Why?What should we be doing?
Warehouses (Relational & OLAP)
Transaction Systems Flat, Legacy or Modern
ExternalSources
The Cognos 8 Performance Management System –Th Pl tf I f t t D t il d A hit tThe Platform Infrastructure Detailed ArchitectureUser Experience Universal CapabilitiesAccess and Interactivity Business Modeling
ContentIT Tools Shared Set of Purpose-Built Services
Common Business Model
Model
Admin
Upgrade
Shared Dimensions
Personal Datasets
Presentation Service
EventService
Security Service
Cognos 8 Bus – Dispatcher SOAP, XML
SchedulingService
Cognos 8 A hit t
Content
Security
Services API
Plan & Report
QueryService
MovementService
Calculation Service
Streaming Service
Attachments, AnnotationsArchitectureInitiatives,
MetricsSystem ContentMetadata, Events
Optimized OLAP64-bit in-Memory
MetadataPlan & Report Saved Objects
Open Data Access
Metadata, Events Search Index, Audit Logs, System Metrics
64 bit in MemoryHigh PerformanceStreaming Cache
Exploit existing infrastructure choices Integrated & optimized with IBM infrastructureI f t t
Modern and Legacy Sources
Application Sources
OLAP Sources
Access to All Data
Message Sources
Relational Sources
p g g p
Data Integration & Data Quality Tools
Platforms & Databases
Security Providers & Firewalls
Application & Web Servers
Infrastructure Fit
The Cognos 8 Platform easily Integrates with Existing Environments
Any J2EE application serverApache Tomcat (incl ) IBM webSphere
Environments (Simultaneously available)
App ServersSOAP, XML, WSDL, HTTP(S), OMG
Interoperable and Extensible
Standards Apache Tomcat (incl.), IBM webSphere, SAP NetWeaver, BEA WebLogic, Oracle Application Server 10G
Any combination of Windows, Unix or LINUX (RedHat)
Operating Systems
App ServersSOAP, XML, WSDL, HTTP(S), OMG CWM, XML/XMI, WSRP,RESTStandards
ExtendOpen, Web Services API to programming language of choice including Java, Microsoft .NET Framework, COM & Perl
Industry Standard Portals including IBM, y
Routers/Firewalls All standard gateways supported
y g ,SAP, Plumtree, WRSP)
Enterprise Search, BPM, Office, 3rd party Applications
Embed
Any combination of LDAP, NTLM, Active Directory, Netegrity, SAP, Cognos 7, customSecurity
Intel, AIX, HP/UX, SolarisPlatforms
Firewalls
Metadata Integration Open Data Access
y, g y, , g ,
Cognos 8 Platform
CA ERwin, Oracle Designer, Business Objects Designer, Sybase PowerDesigner, CA COOL, Rational Rose, XML/XMI, OMG CWM
Modeling tools
Relational
XML, WSDL, LDAP, JDBC, JDBC…
IBM DB2 UDB, Oracle, Microsoft SQL Server, Teradata, other…
Modern
Source catalogs Relational databases, ERP
PowerCubes: High Performance Dimensional CacheI d t St d d OLAP P id
SAP, PeopleSoft, Oracle, Siebel, salesforce…ERP
OLAP
Cognos Data Manager, Informatica PowerCenter, IBM Information Server (Ascential) Metabroker, Composite, IBM II
ETL & EII
gIndustry Standard OLAP Providers: Microsoft, IBM, SAP, Essbase
Short Demonstration of just a little piece f Cof Cognos on z
Results of PoC of Cognos on System z
Beyond demonstrations...
A joint effort of IBM Software Group Belgium, IBM Montpellier and NumiusIBM Montpellier and Numius
Description Objectives and Technical architectureDescription, Objectives and Technical architecture
Mission statement of the joint IBM-Numius project
T f ll t i ti b i i t lli• To successfully port an existing business intelligence environment from an Intel-HP / Windows – HPUX –Oracle 10g – MS SQL server 2000 architecture to anOracle 10g MS SQL server 2000 architecture to an IBM Cognos 8.3 on System z – IBM DB2 9 for z/OS architecture.
• To prove that the end-user functionality is maintained.• To perform tests to determine the performance p p
boundaries of this real-life application on a simple System z hosted architecture.
• To learn what are real-life best practices when executing a “port” of this type.
Client contextClient context
• Numius’s client provides administrative and financial• Numius s client provides administrative and financial (data) services to about 10.000 Belgian companies, mainly in the retail sector.y
• Numius’s client uses IBM Cognos 8 for internal data l i d f lanalysis, process management and for external
communication with its 10.000 B2B clients and with about 200 B2B business partners.about 200 B2B business partners.
Logical architecture at Numius clientLogical architecture at Numius client
Input of data Intermediate OperationalInput of data
Application in Oracle FORMS
Intermediate datasource
Operational datasource
Application in Oracle FORMS
Client on internet Once per night
File based interface(client generates the neccesary files)
Logical architecture continuedLogical architecture - continued
Operational datasource
Data warehouse Powercubes Users
OLAP
ETL
Regeneration of cubes
Internal users
Incremental process, every night External users
RDBMS
OLAP RDBMS
The illusion: big bang for
no bucks
uctiv
itysi
ng p
rod
Incr
eas
Cost level
© Numius nv
The actual client (distributed) Cognos architecture
© Numius nv
Typical Distributed situationTypical Distributed situation
Client challengesClient challenges
• Numius’s client faces strongly degrading performance• Numius s client faces strongly degrading performance, both for database queries as well as for OLAP processing.g
• Numius’s client wants to outsource entire computer infrastructure to a central computing centre.N i ’ li hi i f l b• Numius’s client wants to achieve economies of scale by simplifying the heteregenous distributed computing architecture.architecture.
• Numius’s client wants to use its business intelligence tools as an individualised communication channel with
(its stakeholders (clients, business partners, shareholders,...)
Th ill i The vision:The illusion: big bang for small bucks
The vision: big bang for
correct bucksuc
tivity
The start: asing
pro
d
The start: a small fizz for a few
bucksIncr
eas
Cost level
The distributed environment the client should have
The distributed high availability equivalentThe distributed high availability equivalent
The vision:The illusion: big bang for small bucks
The vision: big bang for
correct bucksuc
tivity
bucks
Th t t Theng p
rodu
The start: a small fizz for a few
bucks
The nightmare: big bucks
but no fi k
Incr
easi
bucks fireworks
Cost level
The PoC z technical architectureThe PoC z technical architecture
Logical infrastructure of the PoC
Client Machine
W b BWeb Browser
HTTPHTTP
Application Server Machine
Application ServerppJVM
Cognos 8 BI Services
DataSQLSQL
RDBMS Server
Sourceson z/OS Content
Store
Test 1: Determine top capacityTest 1: Determine top capacityity
Marginal productivity > 0System “underutilized”
Marginal productivity near 0System at top capacity
Marginal productivity < 0System “overutilized”
Pro
duct
ivi
P
Increasing load
Test 2: Determine end-user comfort zone
mes End-user comfort zone
pons
e tim
ppl. Controller
Injector 1
Injector 2
System Z
user
resp
in w
eb a
p
Injector 3
d-to
-end
pe
rienc
e
Increasing numberf i t ti
End
exp
of interactive users
Performance test structurePerformance test structure
Lane 1 of 1 x 4 reports
Lane 2 of 1 x 4 reports
RDBMS OLAP RDBMS OLAP
p
Lane 3 of 1 x 4 reports
...Lane 100 of 1 x 4 reports
T0 T+2 T+4 T+6 T+8 T+10 T+12T+14 T+n
Explanation of Performance test dynamics• Performance tests environment• Performance tests environment
– During all the tests, the same architecture on System z was used.
– The following parameters where changed to create different test scenarios:
• Number of parallel 1 x 4 lanes started at a single moment in time• Number of parallel 1 x 4 lanes started at a single moment in time.– 1 x 1 x 4– 10 x 1 x 4
21 1 4– 21 x 1 x 4– 51 x 1 x 4– 100 x 1 x 4
Wi h i h /OS WLM Di i S i G l• With or without z/OS WLM Discretionary Service Goal• With or without DB2 9 for z/OS parallelism
Delicate team play between architectural layers
Linux for System z processing power
IBM Cognos layer
IBM Cognos 8.3 Number of Processes tuning for Batch Report Service
IBM DB2 z/OS layer DB2 9 for z/OS processing powery
Phase 1: N x 4 reports are submittedPhase 1: N x 4 reports are submitted
IBM Cognos layer
IBM DB2 z/OS layery
Phase 2A: Simple reports/queries are quickly finished
IBM Cognos layer
IBM DB2 z/OS layery
Phase 2B: Complex reports/queries take longer to process “wait time” in output
IBM Cognos layer
IBM DB2 z/OS layery
Phase 3: Uniform distribution as easy and complex reports are processed randomly
IBM Cognos layer
IBM DB2 z/OS layery
IBM Cognos 8 Report flow for 100 x 1 x 4 test without z/OS WLM Discretionary Service Goal
• The IBM Cognos 8 batch report• The IBM Cognos 8 batch report service is tuned to allow 6 parallel processes, generating a constant processing flowconstant processing flow combined with a steadily decreasing queue of reports that
iti f i l tare waiting for a processing slot to become available (in the IBM Cognos 8 layer).
IBM Cognos
Linux for System z processing power
IBM Cognos layer
IBM DB2 Z/os layer
IBM Cognos 8.3 Number of Processes tuning for Batch Report Service
DB2 9 for z/OS processing power
Linux for System z CPU for 100 x 1 x 4 test /o /OS WLM Discretionar Ser ice Goalw/o z/OS WLM Discretionary Service Goal
• At the Linux for System z level, we see peaks of processing and peaks of I/O
• These peaks are caused by the large PDF reports y g pthat need to be generated.
WLM Workload• WLM = Workload Manager
SQL activities for 100 x 1 x 4 test w/o z/OS WLM Discretionary Service Goal
• At the DB2 9 for z/OS
L100RP01# SQL Activities
1000
1200
level, we also notice an increase in corresponding SQL requests.
400
600
800
SQL
0
200
3:28:00
3:30:00
3:32:00
3:34:00
3:36:00
3:38:00
3:40:00
3:42:00
3:44:00
3:46:00
3:48:00
3:50:00
3:52:00
3:54:00
3:56:00
3:58:00
4:00:00
4:02:00
4:04:00
4:06:00
4:08:00
4:10:00
4:12:00
4:14:00
13:
13:
13:
13:
13:
13:
13:
13:
13:
13:
13:
13:
13:
13:
13:
13:
14:
14:
14:
14:
14:
14:
14:
14:
DB2 9 for z/OS CPU for 100 x 1 x 4 test comparison
L100PR01# CPU Utilization 46 mins
• 6 CPs = 100% blue line• 1 zIIP = 100% pink line• IIPCP yellow line 150 00%
200.00%
250.00%
300.00%
CPU-Busy%zIIP-Busy• IIPCP yellow line
e g first peak approx. 250% equates to additional 2.5 zIIP 0.00%
50.00%
100.00%
150.00%
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
zIIP BusyIIPCP%
additional 2.5 zIIP capacity
• We notice a big difference in elapsed time
13:28
:0013
:30:00
13:32
:0013
:34:00
13:36
:0013
:38:00
13:40
:0013
:42:00
13:44
:0013
:46:00
13:48
:0013
:50:00
13:52
:0013
:54:00
13:56
:0013
:58:00
14:00
:0014
:02:00
14:04
:0014
:06:00
14:08
:0014
:10:00
14:12
:0014
:14:00
104 mins
in elapsed time• Total CPU time is
equivalent in both runs
Comparison of Linux for System z CPU with and without z/OS WLM Discretionary Serviceand without z/OS WLM Discretionary Service Goal
With t WithWithout With
• Higher processing power of DB2 9 for z/OS generates more intensive
• Lower processing power in DB2 9 for z/OS leads to longer elapse time in g
peak load on Linux for System z. • If the OLAP processing – which is entirely handled on the Linux for
g pquerry resolution. • It translate in a more uniform load distribution on Linux for System z, and y
System z – coincides with one of these peaks, then the OLAP processing time will go up.
y ,no impact on OLAP processing.
g g
Comparison of results with and without z/OS WLM Discretionary Service Goal
With t WithWithout With
M h h t t t l d ti• Much shorter total duration• Shorter “silence” in beginning of test
Comparison of query execution time with and without z/OS WLM Discretionary Servicewithout z/OS WLM Discretionary Service Goal
With t WithWithout With
• Significantly shorter query processing times.
CPU distribution over zIIP / IIPCP / CPCPU distribution over zIIP / IIPCP / CP
With t WithWithout With
Performance Test Phase 2 zIIP eligibility in %
100%
Performance Test Phase 1 zIIP eligibility in %
100%
40%
50%
60%
70%
80%
90%
100%
zIIP%zIIP-CP%CP%
50%
60%
70%
80%
90%
100%
zIIP%zIIP-CP%CP%
0%
10%
20%
30%
40% CP%
zIIP% 53.5% 19.6% 20.0% 20.0% 24.3% 20.1%
L01RP01 L10RP01 L21RP01 L51RP01 L51RP02* L100RP01 0%
10%
20%
30%
40% CP%
zIIP% 55 0% 53 8% 54 4% 76 2% 76 4% 54 4%
L01RP01 L10RP01 L21RP01 L21RP02* L51RP01* L100RP01
• zIIp-CP processing takes up a substantially larger proportion
• zIIPCP processing only for none discretionnary periode
zIIP-CP% 1.5% 35.4% 35.0% 35.0% 53.6% 34.9%
CP% 45.0% 45.0% 45.0% 45.0% 22.0% 45.0%
zIIP% 55.0% 53.8% 54.4% 76.2% 76.4% 54.4%
zIIP-CP% 0.0% 0.0% 0.0% 1.7% 1.5% 0.5%
CP% 45.0% 46.2% 45.6% 22.1% 22.1% 45.0%
substantially larger proportion. discretionnary periode
Teamplay of different layersTeamplay of different layers
• We notice how the production of complex reports on complex queries• We notice how the production of complex reports on complex queries requires coordinated teamplay between the different layers of the architecture.Q er resol tion is follo ed b PDF generation hich in t rn res lts in• Query resolution is followed by PDF generation which in turn results in report delivery to the end-user.
L100RP01# SQL Activities
600
800
1000
1200
SQL
0
200
400
13:28
:0013
:30:00
13:32
:0013
:34:00
13:36
:0013
:38:00
13:40
:0013
:42:00
13:44
:0013
:46:00
13:48
:0013
:50:00
13:52
:0013
:54:00
13:56
:0013
:58:00
14:00
:0014
:02:00
14:04
:0014
:06:00
14:08
:0014
:10:00
14:12
:0014
:14:00
Conclusions on performance testsConclusions on performance tests
• The platform behaves in a stable and predictable• The platform behaves in a stable and predictable manner.
• System behavior is driven strongly by the report types.System behavior is driven strongly by the report types. Architecture behaves in function of usage.
• A good architecture is “TeamPlay”. The teamplay must b d h f h l fbe geared to the usage type of the platform.
• All components must be tuned appropriately.A lti ti IBM C 8 hit t i l t• A multi-tier IBM Cognos 8 architecture remains relevant, but System z can significantly reduce TOC.
• A SINGLE 1x4 lane on the Numius customer systemA SINGLE 1x4 lane on the Numius customer system took approximately 45 minutes of processing time.
Results of Scaleability Test on a IBM DB2 9 for z/OS RDBMS datasource
Li f S t i h d h l l b i thi• Linux for System z is charged much more regularly because in this end-to-end test, interactive end-user behavior must be accomodated too by the Linux layer.
Results of Scaleability Test on an IBM DB2 9 for z/OS RDBMS datasource
• The test ran successfully for the full 130 virtual concurrent users that were included in the testthe test.
• As of 114 users we experience slower “surfing” behavior.• We learnt that this was caused by the fact that we artificially limited each virtual
user’s bandwith, thereby unnecessarily “occupying” processing power of the Linux for System z at larger user numbersSystem z at larger user numbers.
• In real life, this number of virtual users should be multiplied by 10 to 50 to achieve a number of potential “named” users.
• On the client’s system, this test could only be run for 8-12 users.
Conclusions from end-user perspective• Numius, as a representative to the client, can conclude , p ,
the following:
– The application was successfully and without loss ofThe application was successfully and without loss of functionality ported to the System z platform. This required no redevelopment.
– The client’s application would not require a redesign to pp q gaccomodate its growth in data volumes or in terms of users.
– Reports that are not practically useable at client’s site now become relevant again. Reports that did not run at client’s site
blnow are runable.– Client would be able to serve many multiples of current
number of users with the very simple architecture from this PoCPoC.
– Client could scale out to more complex architecture without increased hardware complexity.
The illusion:
The z: big bang for correctThe illusion:
big bang for small bucks
correct bucks
duct
ivity
The start: a The nightmare:si
ng p
rod
e sta t asmall fizz for a few
bucks
nightmare: big bucks
but no fireworks
Incr
eas
fireworks
C t l lCost level
Contact us atContact us at
J C tJo Coutuer+ 32 478 20 40 18 32 478 20 40 [email protected]
NumiusNumiusInterleuvenlaan 1515D3001 Heverlee