STAC Report: Near Real-Time Position & Risk Management for a Large Set of Positions (GemFire 6.3 on HP blades)

HP and Gemstone show 76x improvement in response times by using "process and data affinity" within a data grid STAC has prepared a STAC Report entitled: “Near Real-Time Position & Risk Management for a Large Set of Books: GemFire 6.3 on HP BL460c Blades and Voltaire InfiniBand” HP discussed this benchmark at the June 21 STAC Performance Summit in New York (click here for the slides and video). HP and Gemstone constructed a system to demonstrate the value of process and data affinity—i.e., performing calculations as close to data as possible. The vendors believe that achieving near real-time analytics for risk management requires a new architecture that holds all information in memory with fast access to expansive compute. The vendors wished to demonstrate a “data aware” architecture that can support a large set of positions and still outperform the more traditional “data unaware” architecture. The use case they chose was position keeping on a set of equity options. The tests subjected the system to a realistic workload and compared the response times for end-user position requests when the system was configured two ways: 1. A traditional data grid, in which position calculations are performed on the compute-grid nodes 2. A configuration exploiting new GemFire features that enable calculations to be performed within the data grid. The system responded to requests for positions on 20 books of 10,000 options each, starting with 5 million trades per book, with 20,000 market data updates per second and 2,000 new trades per second. STAC verified that the system was resilient to server failure. Using the new data-aware execution service in GemFire 6.3, the mean time to reprice books was 76 times faster than the data-unaware GemFire configuration. Click here for the full STAC Report.