Writeimage CSB Reinsurance Risk Experts Cut Catastrophe Modelling Time by up to 75 Per

“We’re at the forefront of catastrophe modelling with our platform based on Windows HPC Server 2008. For example, due to the increased performance we can run a particular flood model in one hour instead of four.”

Nigel Davis, Head of Platforms & Delivery, Product Development, Willis Re

Willis Re―supported by its research arm the Willis Research Network (WRN)―runs increasingly complex natural catastrophe risk models to keep pace with developments in its field. To meet the challenge, the Willis Re Analytics Product Development Team created a high-performance computing (HPC) environment built on Windows HPC Server 2008. The organisation can now run selected models in a quarter of the time, delivering a faster response to client needs.

Business Needs

Willis Re, which is part of the Willis Group, is one of the world’s leading reinsurance advisors. It helps organisations understand the risks they face and advises them on the best way to manage extreme events. Willis Re supports and funds the WRN—a collaboration between universities and the insurance industry—to evaluate the frequency, severity, and impact of natural catastrophes, and help governments and businesses manage risks more effectively.

At the heart of Willis catastrophe modelling activities is an integrated, global catastrophe modelling platform called eQUIP. This platform hosts catastrophe models covering many countries and calculates the effects from various natural and man-made hazards. The modelling of client data—which can cover millions of geographical locations—requires thousands of simulations to estimate the potential financial loss.

With the science behind catastrophe modelling improving, the amount of data expanding, and time pressures from clients increasing, Willis wanted to raise the platform’s performance. Hélène Galy, Head of Modelling Product Development at Willis Re, says: “We wanted eQUIP to be among the most powerful catastrophe modelling platforms in the insurance market, featuring the latest HPC technology.”

The project carried significant risks because the complex software behind the catastrophe models required redeveloping for a HPC environment. Nigel Davis, Head of Platforms & Delivery, Product Development at Willis Re, says: “We were breaking new ground in catastrophe modelling.”

Work also had to follow a strict schedule to meet the insurance industry’s busier renewal periods at the end of March. It was crucial for Willis to find an IT solutions provider with experience in HPC technology, and the resources to support the software redevelopment cycle to keep work on track. “We had set ourselves some challenging goals, but we felt certain they were achievable,” says Davis.

Solution

Willis turned to Microsoft, which had been its strategic partner for a number of years. Davis says: “We chose Microsoft because it really understood our business and our aims. In addition, Microsoft had the capabilities to deliver an end-to-end solution.” Davis was impressed with the response of the Microsoft U.K. team. “It dedicated a lot of time to understanding our business and what we wanted to achieve,” he says. After a series of technical discussions, Willis decided to migrate the eQUIP platform to a HPC architecture based on Windows HPC Server 2008, with software written in Microsoft .NET.

The design for the HPC environment was simple. It included eight servers and 64 central processing units (CPUs) providing the computing performance for parallel processing, which allows applications to be processed by multiple CPUs at the same time. However, ensuring the models ran well on a HPC architecture was more complex. Davis says: “Microsoft provided us with a lot of support during the development and testing of our models.” A Microsoft partner in Germany checked the modelling software on a similar HPC grid. “We undertook extensive testing and worked with Microsoft technicians to optimise the code. It was great to gain access to their expertise,” says Davis.

In January 2011, Willis tested the HPC platform internally with the aim of running it on behalf of clients at the end of March. Galy says: “We’re now loading all the catastrophe models onto the platform. These models cover everything from floods in Europe to windstorms in Asia, and we’re running the final checks—it’s gone very well.” Based on this success, Willis will be developing the eQUIP platform’s functionality to create a web interface for clients to access modelling results online. It then plans to integrate eQUIP with its geospatial platform.

Benefits

With the eQUIP platform, Willis can maintain its leading position for catastrophe modelling and continue meeting the need for more detailed models. Davis says: “We’re at the forefront of catastrophe modelling with our platform based on Windows HPC Server 2008. For example, due to the increased performance we can run a particular flood model in one hour instead of four.”

• Parallel processing ensures faster modelling to meet growing demand. The HPC architecture has increased the eQUIP platform’s computational power because jobs are shared between the solution’s multiple CPUs. “Thanks to the increased speed of our platform, we can meet the growing demands for catastrophe modelling worldwide,” says Davis.
• More detailed models deliver improved service to clients.The platform’s performance means Willis can create models that deliver more accurate information. Davis says: “We can take advantage of improvements in the quality of data and build more complex models safe in the knowledge that our platform can meet the processing requirements.”
• Great scalability supports easier expansion. As the science behind catastrophe modelling develops and the quality of data improves, there will be a need for even greater computing power. Willis can take advantage of these opportunities because its platform scales easily thanks to the simplified administration. “We’ve created a highly flexible infrastructure that we can expand quickly with Windows HPC Server to meet the growing computational needs of catastrophe modelling,” says Davis.
• Simple design creates repeatable model for Willis business units. After the success of the eQUIP project, other tools within Willis Analytics will be considered for migration to HPC. Davis says: “We are excited that the same HPC architecture will be supporting other analytical resources in the near future—it’s game changing.”