Large U.S. based Proprietary Trading Firm - Customer Case Study

A large U.S. proprietary trading firm, specializing in high frequency trading, recognized early on that they needed to move away from their direct attached and NAS storage systems. This architecture didn’t allow them to access and share petabytes of data across multiple types of high-performance trading groups, including: teams in currencies, derivatives, international equities and technology equities. An extreme need for speed also was identified, since our customer’s success and competitive advantage depended upon bringing new strategies to bear quickly (while phasing out unsuccessful ones) in real-time (or as close as possible). A parallel architecture was selected providing two key criteria to meet these goals: 1) a global namespace for more efficient data gathering and sharing, and 2) parallel I/O to remove the latency associated with sequential jobs running on traditional NAS architectures. Several generations of parallel infrastructure were tested, based on storage and parallel file system, including offerings from major server and storage vendors. However, in production conditions, neither could meet the minimum performance requirements for current and projected needs, which are in the petabytes and GB/s of sustained I/O performance. The firm was impressed with DDN’s massively parallel architecture and open platform approach to parallel file systems, which would permit infrastructure changes without forcing them to replace their entire investment. The company evaluated and then selected the GRIDScaler™ appliance, which tightly integrates an enterprise parallel file system with high performance SFA storage. This solution which surpassed their performance, availability, scalability and cost requirements. The CIO also liked DDN’s Python-based infrastructure and features, which would support possible future directions, such as embedding applications right inside the storage, cutting the I/O path process steps by nearly half to deliver much lower latency.