Risk Modeling with Condor at The Hartford

1
Risk Modeling with Condor at The Hartford

• Condor Week
• March 15, 2005

Bob Nordlund The Hartford robert.nordlund_at_hartford
life.com
2
About The Hartford

• Almost 200 years old (founded in 1810)
• 18 billion 2003 revenues
• 250 billion assets under management
• 30,000 employees worldwide
• Diverse products
• Property Casualty Insurance
• Individual and Group Life Insurance
• Group Benefits
• Individual Annuities
• Not a technology company!

3
The Need for Speed

• Changes in the reinsurance market necessitated an
  alternate approach to limiting our exposure -
  hedging.
• In order to accurately hedge our book of business
  we needed to fully understand our liability
  position.
• Current methods of liability modeling at The
  Hartford were inadequate to satisfy our auditors.
• We needed more power much more power, and
  quickly.

4
Band-Aids

• Extend the master/worker capabilities of the
  current modeling tool.
• Buy dedicated server-class hardware
• Improve networking
• Create policies and procedures to improve
  reliability
• Results
• Largest distributed implementation of modeling
  tool in the world
• Tested to 68 workers
• Still not good enough.
• Reliability concerns
• Breaking new ground for scalability
• Manually intensive operation

5
The Promise of Grid

• In Q2 of 2003, a commercial grid provider
  approached us with an integrated solution for our
  modeling tool.
• Infinite scalability
• Full recoverability
• Integrated
• Not a production release
• Very expensive
• At the same time, we hedged our technology bet
  with an in-house solution using Condor.
• Scalable
• Mature
• Flexible
• Free
• Supports Windows

6
And the Winner is Condor

• The commercial solution was not delivered to our
  satisfaction.
• Condor met every expectation for performance,
  reliability, scalability, and ease of use.
• Condor had excellent support for Windows.
• Condor licensing was flexible and affordable.

7
Pitching Condor

• Free software is not always an easy sale.
• FUD from vendors can be very effective.
• Who do you blame?
• Where do you get support?
• Who can I call?
• Mitigating Factors
• The community on the condor-user list provided
  better support than most vendor support options.
• UW offered paid support options.
• Our prototype implementation worked.
• Potential cost savings could not be ignored.

8
Condor at The Hartford

• Initial production hedging implementation in Q3
  of 2004.
• Grown from 12 dedicated HP DL360 G3 2 processor
  servers to 200.
• Added a secondary pool for non-production
  analytical jobs.
• Produces and analyzes 4TB of data per month.
• Runs 100K analytical jobs per month.

9
Daily Production Hedging Jobs
10
Daily Analytical Jobs
11
Future Directions

• Improve our Condor environment.
• End-user automation
• Multiple pools
• Improved security
• Better policies
• Accounting
• Extend our use of Condor to non-dedicated
  hardware.
• 1,000 file servers unused or underutilized after
  business hours
• 800 application and database servers
• 25,000 desktops
• Create new modeling applications to exploit
  Condor features.
• Master/Worker
• Checkpointing
• Find new applications for Condor throughout the
  company.

12
What Wed Like in Condor

• Hot failover capability for all Condor subsystems
• Integration strategy for alerting and monitoring
  frameworks for both daemons and jobs.
• Simplified end-user job submission and
  monitoring.
• Graphical management console

13
In Closing

• Condor is a terrific technology that will only
  grow in size and importance at The Hartford.
• It has enabled unparalleled productivity gains in
  area of risk modeling.
• As more applications for the technology are
  identified more devoted users are surely to
  follow.

14
Thank you.
Bob Nordlund The Hartford robert.nordlund_at_hartford
life.com