Case Study

1
Lecture 14

• Case Study
• Postgirot Bank and Provment AB

2
Case Issues

• Industrial Engineering approach to managing
  corporate server capacity
• Software measurement systems used to keep track
  of how various components of mainframe
  configuration are being used
• Systems try to manage risk of over or under
  capacity
• Case describes new methodology for applying this
  approach to servers rather than software

3
Postgirot

• Swedish-based company providing large, cashless
  settlement system
• Technology is core
• Potentially for sale and has strong need to be
  seen as efficient to reduce job turnover
  post-merger
• IT strategic in handling 1 million transactions
  per day
• Should be at leading edge of development and
  operating practice
• Leading-edge developer of IT with a massive
  effort to webify entire company
• Located in very IT-friendly country
• Must be available 24/7
• Leading edge measurement tools is good move

4
Postgirot IT unit

• 400 people
• 74m budget
• Unevenness in workflow
• Peak hourly volumes four times average hour
• Deeply dependent on reliability
• Built on underground information-processing
  center to hold central computer and server farm
• 158 staff focused primarily on development of
  capabilities to support organization
• Cost of services historically of secondary
  concern to getting capabilities installed
• Culture of speed and effectiveness instead
• Stage 2 contagion company on Nolan scale
• Substantial excess capacity discovered on servers
  with ability to cut costs around 5 without
  harming productivity

5
Provment A-B

• Staff has had significant experience developing
  software for measuring mainframe performance in
  OS/390 environments
• Recently expanded tool to UNIX systems servers
• New twist on old theme, relying on industrial
  engineering tools in operations environment
• Driving excess costs out of network without
  degrading server is key
• Recently developed new software package to
  measure the utilization of server capacity
• Complex CPU performance measurement system,
  internal memory utilization, I/O capacity and
  disk storage use
• Measures for each server how many devices are
  being utilized
• Unobtrusive, takes up less than 0.5 of CPU
  capacity
• Package hasnt been validated in this form yet

6
Key aspects of new package

• Intensely secure
• Tightly controlled user system access
• Designed to keep servers running if disabled
  (even though this is not anticipated to happen)
• All staff for installment provided by Provment as
  a service
• Eliminating training for Postgirot staff
• Lowering barrier for first trial
• Maximum time to install on all servers 60 days to
  ensure rapid access by management to results
• All installation costs added together not more
  than cost of 4 UNIX servers

7
End results for UNIX servers

• Discovered average workload of first 48 servers
  less than 10 during prime shift
• Up to 40 servers could be eliminated
• Not unexpected
• Machine consolidations expect to save 500k in one
  time, eliminating capital investment and 1.4m
  per year (5 of IT budget)
• Each server on full-cost basis about 3 times more
  expensive than anticipated
• Management didnt realize cost of what they wer
  running
• Can they generate these savings?
• System showed high potential for NUNIX servers
  being combined
• NT/WIN servers suggested as next step

8
Issues

• Quality of analysis in establishing cost
  coefficients new area/idea
• Potential misuse of Postgirot data by Provment
• How sensitive will system be to peaks?
• Betting important measurement function on new
  company?
• No mention of applications, which are sensitive
  to down time and should be on reliable mainframes
• Concern that these issues are addressed piecemeal
  than holistic