ActiveActive in the Real World

1
Active/Activein theReal World

• Neil Barnes
• Technical Specialist
• November 12, 2008

2
Contents

• Who am I and who do I work for.
• What do I mean by Active/Active.
• Why did we go to Active/Active.
• How do we do Active/Active.
• What happens when we go single node.
• Benefits and Issues
• The Future

3
Who am I and who do I work for?

• Neil Barnes Senior Technical Specialist at Lloyds
  TSB plc
• Lloyds TSB is fifth Largest retail Bank in the
  United Kingdom
• Founded in 1765
• I have worked on Tandem systems for 28 years,
  last 22 for Lloyds TSB
• I am Vice Chairman of British Isles Tandem User
  Group (BITUG)

4
CONTINGENCY
CONTINGENCY
ACTIVE PRIMARY

• ACTIVE PRIMARY

5
(No Transcript)
6
Original Contingency Configuration
7
Active/ Active Contingency Configuration
50 Redundancy
BASE24-atm LN 2
BASE24-atm LN 1
BASE24-atm LN 1
BASE24-atm LN 2
AB Mode
2000 ATMs
2000 ATMs
8
Contingency Switch
2000 ATMs
9
File Transfers and Replication
In-house Transfer and Replication Routines
10
File Transfers and Replication (cont.)
Third Party Replication Solution
DRNet
11
CAF

• Contains details of Cards
• Made up of 35 Million Records
• Updated when a transaction is performed on the
  card
• Each card is unique

12
(M)TLF

• A subset of the Transaction Log File
• Keeps a daily track of all financial transactions

13
IDF

• Contains institution details
• Only updated when we do a full CAF refresh

14
ISECD

• Holds encryption keys for exchange with
  interfaces (e.g. LINK)
• Replication is via the TMF Audit trails
• Updates are via the logical networks

15
CSECD

• Holds the encryption keys used when talking to
  the ATMs
• Same replication as the ISECD

16
File Transfers and Replication (cont.)
ACI Replication Solutions
FHM
REFRESH
Hosts
17
Major Benefits

• Minimizes unexpected problems on contingency
  system
• Improved system resilience
• Reduces Tandem load

18
Minimizes unexpected problems on contingency
system

• The system is accessing all the NonStop Hardware
• All Communication lines are being used
• We know if we have a component failure

19
Improve system resilience

• Communication lines are at 50 capacity
• A loss of a CPU has less impact on the system

20
Reduce Tandem Load

• CPUs have less transactions going through
• Communication lines have half the traffic
• Less disk accesses

21
Issues

• Complex system management/support tools required
• Down-time still required for any contingency
  switch
• System sizing

22
Complex system management/support tools
required

• An extra layer is added to infrastructure
• Management tools have to be able to run on one
  system and access/change details on another.

23
Down-time still required for any contingency
switch
Tandem A
Tandem B
BASE24-atm LN 1
BASE24-atm LN 2
BASE24-atm LN 2
BASE24-atm LN 1
24
System sizing

• We only switch to single node on a quiet day
• Performance data taken on single node running
• Performance data also taken on peak day
• What would a peak day running single node look
  like?

25
The Future

• Remove dependence on in-house file transfer
  replication routines
• Extend online replication
• Reduce the reconfiguration required on any
  contingency switch
• Itanium

26
Remove dependence on in-house file transfer
replication routines

• Some of the replication routines are written in
  house
• Costly to maintain especially at upgrade time
• Support staff forget how things work.

27
Extend online replication

• Move more files to use DRNet

28
Reduce the reconfiguration required on any
contingency switch
Tandem A
Tandem B
BASE24-atm LN 1
BASE24-atm LN 2
BASE24-atm LN 2
BASE24-atm LN 1
BASE24-atm LN 1
29
Itanium

• Looking at a hardware refresh in the next couple
  of years
• No longer need Active/Active for Communication
  failures
• Fast Failover may be better than what we have now
• Single node is easier to manage
• Systems now so reliable that the Active/Active
  may not be cost effective

30
Summary

• Active/Active has been used by Lloyds TSB for at
  least 15 years
• Half our ATMs are available during a switch
• The performance and capacity functions need to
  look at peak day when running in single node

31
Thank You
Any Questions?
32

• Neil Barnes
• Lloyds TSB plc
• Neil.P.Barnes_at_lloydstsb.co.uk