Kiwibank: Startup to Enterprise in 7 years
Wayne Knowles
Ki ib
k t St
t
2
Kiwibank at Startup
Established by Labour-Alliance Coalition Government
Low Startup BudgetLow Startup Budget
Leveraged NZ Post for Infrastructure
Branch Network using existing Postshop’s and franchiseBranch Network using existing Postshop s and franchise network
Leverages ISP Services
Ki ib
k t St
t
2002
3
Kiwibank at Startup - 2002
IBM p-Series for “Core” banking database
IBM x-Series machines running Microsoft Applications
File & Printer services
E h M il
Exchange Mail server Citrix
Several SQL Server databases
Entry Level IBM SAN
Fil & P i t l SQL S D t b (< 1TB) File & Print plus SQL Sever Databases (< 1TB)
R
d T S
4Road To Success
2002 2003-2008 2009 Launch • Project team 2003-2008 • Strong growth 2009 • Success! j • No customers g g• Acquisitions • Many customers
• Guess
requirements
Not sure what we
• Brand developed
• Building market knowledge
• Many products
• Know what customers expect
• Not sure what we would become!
Ki ib
k G
th
5
Kiwibank Growth
U d
t
d Th A
li
ti
6
Understand The Application
Understand how CPU, memory and storage is impacted by growth
- Linear: resource_usage = n x customers
Exponential: resource_usage = n x customers2
Silicon is faster than Disk (Ferrite):
Keep “working set” in memory for 24 hours
Expect high I/O and performance impact after a reboot
U d
t
d Th A
li
ti
7
Understand The Application
Eliminate the high cost operations: Disk I/O ReadsDisk I/O Reads
Process startups (forks)
Best way to find bottlenecks: perform full load test
Commercial tools available (Tivoli Loadrunner etc)
Commercial tools available (Tivoli, Loadrunner etc)
Use Standard OS Tools (truss, sar and perfmon)
Especially when sharing performance data with external vendorsEspecially when sharing performance data with external vendors
U d
t
d St
G
th
8
Understand Storage Growth
Data can be classified into 2 categories: StructuredStructured
Unstructured
Most organizations recognize that unstructured data Most organizations recognize that unstructured data
growth is out of control.
Multiple copies of the structured data can also result in a large storage requirement.
St
G
th
9Storage Growth
UAT UAT PROD DR UAT UATStage MIS ODS
Consider the storage requirements of multiple test g
environments!
A hi i d D D h l thi th
S
t C
it M
t
10
Smart Capacity Management
“Radar Screen” forcasting
One should understand how the application growsOne should understand how the application grows Track the upper limitTrack the upper limit
Ie. xx Months before next impact
When highlighting issues to management use correctWhen highlighting issues to management use correct terminology
Understand Vendor Lead-Times
B ildi
M
bl I f
t
t
11
Building Manageable Infrastructure
Engineer upfront for cater for future upgrades
CPU or Memory upgrades are easy
CPU or Memory upgrades are easy
Capacity on Demand
Consider future Forklift upgrades
Put the Customer first and minimize downtime.
Smart choice of technology to minimize downtime
Kiwibank has performed 2 major forklift upgrades
Bi B
U
d
12
Big Bang Upgrades
Triggered by Batch Window timeTriggered by Batch Window time
Migrated to new machine using mksysb migration (AIX)
Implemented in p-Series LPARImplemented in p Series LPAR
Resources allocated to allow future AIX upgrades
Outsourced the solution – implemented in 8 weeks
Outsourced the solution implemented in 8 weeks
R li bilit
13
Reliability
Measure and improve reliability
Full analysis of all problemsu a a ys s o a p ob e s
Concentrate on repeat problems first
More Fingers = More Problems
Reduce amount of touching on the system to a minimumg y
Automated monitoring and reporting to eliminate daily
checks
Wi k d P
bl
14
Wicked Problems
"Wicked problem" is a phrase used in social planning to describe a problem that is difficult or impossible to solve because of incomplete, contradictory, and changing
requirements that are often difficult to recognize.
Moreover because of complex interdependencies the Moreover, because of complex interdependencies, the effort to solve one aspect of a wicked problem may
reveal or create other problems.p
F t
P
f I f
t
t
15
Future Proof Infrastructure
Thin Provisioning
Virtualization (VMWare, LPAR etc.)Virtualization (VMWare, LPAR etc.)
De-dup
RedundancyRedundancy
In place scale-out and scale up
Aim for zero outage upgrades & changes
B ildi
M
bl A
li
ti
16
Building Manageable Applications
Use established industry standards, where possible
SOA Architecture & XML MessagesSOA Architecture & XML Messages
Use Message Queuing
Implement solutions that are Horizontal & VerticallyImplement solutions that are Horizontal & Vertically scalable
Break into small manageable componentsg p
B
i
C
ti
it
17
Business Continuity
Consider sharing with pre-production environment
DR hardware can used as a pre-production environment a d a e ca used as a p e p oduc o e o e for testing and production support.
Visualize with VMWare / LPAR
Implement replication as high up the stack as possible
Ie. Oracle Data-guard, SQL Server Replication
Can be implemented (or tested) in stages with minimum
BCP / DR
Cl
i A
h
18 PRODUCTIONBCP / DR – Classic Approach
Middleware Internet Banking Core Banking Database EFTPOS & ATM Database DR Middleware Internet Banking Banking Core Banking Database EFTPOS & ATM ATMBCP / DR
Cl
i A
h
19 PRODUCTIONBCP / DR – Classic Approach
Middleware Internet Banking Core Banking Database EFTPOS & ATM Database DR Middleware Internet Banking Banking Core Banking Database EFTPOS & ATM ATMBCP / DR Fi
G
i
d
20
BCP / DR Fine Grained
Kiwibank uses a “Fine Grained” DR strategy
Any system can be separately DR’edy y p y
Separate DR procedure for each component
Kiwibank has future plans to use active-active where ibl
BCP / DR
Fi
G
i
d
21 PRODUCTIONBCP / DR – Fine Grained
Middleware Internet Banking Core Banking Database EFTPOS & ATM Database DR Middleware Internet Banking Banking Core Banking Database EFTPOS & ATM ATMBCP / DR
Fi
G
i
d
22 PRODUCTIONBCP / DR – Fine Grained
Middleware Internet Banking Core Banking Database EFTPOS & ATM Database DR Middleware Internet Banking Banking Core Banking Database EFTPOS & ATM ATMBCP / DR
Fi
G
i
d
23 PRODUCTIONBCP / DR – Fine Grained
Middleware Internet Banking Core Banking Database EFTPOS & ATM Database DR Middleware Internet Banking Banking Core Banking Database EFTPOS & ATM ATMKi ib
k C
t & F t
24
Kiwibank Current & Future
Has completed several upgrades due to customer growth
Agile and able to react to change
Significant investment in IBM x-Series and p-Series hardware
S
25
Summary
Understand your applications Think Smarter, not Harder
Maximize the use of your assets
Adopt new technology and use it to grow your business Minimize outages and Customer Impact
26