DBS Bank's Transformation Journey with AWS and Intel

Overview

• This session discussed the successful partnership between DBS Bank, AWS, and Intel in running a high-performance computing (HPC) solution for DBS Bank's financial risk management and pricing engine.
• The session covered the business challenges faced by the financial services industry, how DBS Bank developed a cloud-native pricing engine (QPE) to address these challenges, and the key technologies and optimizations leveraged from AWS and Intel.

DBS Bank and the Quan and Tech Team

• DBS Bank is the largest bank in Southeast Asia, known for its digital transformation efforts.
• The Quan and Tech team at DBS Bank builds pricing models and engines to enable and scale up the bank's trading business.

Business Challenges and the Need for HPC

• Pricing of financial products requires complex calculations and simulations, such as Monte Carlo simulations.
• DBS Bank needs to perform hundreds of millions of pricing calculations daily for customer pricing, trader revaluations, and risk management.
• The pricing workload is highly parallelizable and dynamic, making it suitable for cloud-based HPC.
• DBS Bank's on-premise infrastructure and legacy pricing engines faced limitations in terms of scalability, performance, and time-to-market for new products.

DBS Bank's Solution: QPE on AWS

• DBS Bank developed a cloud-native Pricing Engine (QPE) using AWS services and Intel technologies.
• Key features of QPE:
  • 100% in-house developed, leveraging existing models and integrating with trading systems
  • Highly flexible and scalable, able to perform 100-200 million pricings per day
  • Utilizes GPU acceleration to achieve 500x performance improvements for certain pricing tasks
  • Covers a wide range of trading products, including FX, Equity, Credit, etc.

Architectural Design and Optimizations

• DBS Bank employed various architectural and optimization techniques:
  • Leveraged AWS services like Amazon ECS, Amazon ElastiCache, and others
  • Implemented a robust, resilient framework with no production downtime since 2019
  • Optimized CPU, memory, and network performance through tools like Intel C++ Compiler and profiling
  • Utilized different scaling methods (CPU utilization, time-based, fixed capacity) based on workload characteristics
  • Optimized data loading performance using the Redis-based caching solution

Performance and Business Impact

• DBS Bank achieved significant performance improvements:
  • 100% CPU utilization on the HPC cluster
  • Reduced IQF pricing time from 1 second to 0.1 seconds
  • Processed 100 million risk revaluations daily with less than 2% overhead
• QPE also helped DBS Bank:
  • Avoid vendor licensing costs by building an in-house solution
  • Reduce time-to-market for new products
  • Improve customer satisfaction and strengthen market position

AWS and Intel Partnership

• AWS and Intel have a long-standing partnership, with Intel powering a significant portion of the instances on AWS.
• Intel works closely with AWS to optimize their latest processor and software technologies for the cloud.
• The DBS Bank team collaborated with both AWS and Intel to further optimize the QPE application and leverage the latest Intel CPU and GPU technologies.

Conclusion

• DBS Bank's successful transformation journey with QPE on AWS and Intel technologies showcases the benefits of leveraging cloud-based HPC for financial risk management and pricing workloads.
• The collaboration between DBS Bank, AWS, and Intel enabled significant performance improvements, cost savings, and business impact.
• The session highlights the importance of end-to-end optimization across the application, compute, storage, and network layers to maximize the benefits of cloud-based HPC.