By Richio Aikawa, Director Vertical Solutions – Financial Services, Panasas. After beginning his career in predictive modeling using Kalman Filtering, Richio Aikawa has spent the 20 years working with enterprise storage and low-latency networking technologies.
The financial services industry delivers leading-edge quantitative trading and risk management strategies using state-of-the-art computational tools and techniques including mathematical and statistical modeling, machine learning, and big data analytics. Some firms have deployed thousands of compute nodes and petabytes of storage in high performance computing (HPC) data centers to quench the constant need for time-constrained testing. Today, those resources are under increasing strain from a legion of talented quant teams concurrently demanding fast, shared access to the extensive and ever-growing datasets. As a result, fundamental speed-accuracy tradeoffs are regularly made in search of optimum profitability.
For predictive algorithmic trading strategies, the complex relationships between the market’s parameters are artfully modeled and the models are validated – backtested – using historical data to see how well each strategy performs. These Monte Carlo simulations are iterated hundreds of thousands of times, generating results that converge to profitability estimates and their likelihoods. Similarly, for risk management, potential loss to trading strategies can be calculated applying the commonly used technique Value-at-Risk (VaR) with Monte Carlo simulations to estimate the maximum loss possible at a probability (or confidence) level over a given holding period.
To ensure timely completion of simulations, where executing 100K Monte Carlo simulation iterations can take several hours, high-performance storage that excels at adapting to demanding, parallel, mixed-sized workloads while still being manageable, available, reliable, and scalable is key.
Unstopping the proverbial I/O bottleneck can allay more than simulation completion times. It can provide an improved foundation allowing re-examination of constraints that necessitated model order reduction (MOR). Comprehensive, optimal higher order financial modeling becomes more viable, enabling increased accuracy and precision, more resilience to unanticipated market movements, and even decreased time to convergence.
One solution to the I/O bottleneck dilemma is a scalable, parallel, data storage system that provides excellent price-performance in a balanced architecture designed with financial services workloads in mind. The ActiveStor® Ultra solution running PanFS is Panasas’ latest and most adaptable, total-performance HPC storage system. It combines a distributed and clustered parallel file system, volume manager, and software-based erasure-coding data protection into an integrated appliance platform. PanFS delivers predictable high performance as a result of automatically adapting to the mixed workloads.
The key to adapting to mixed workloads is a unique feature of PanFS called Dynamic Data Acceleration that eliminates the complexity, fragility, and manual intervention of tiered HPC storage. It uses a carefully balanced set of DRAM for caching, extremely low-latency persistent memory, low-latency NVMe SSDs, cost-effective SATA SSDs, and bulk HDDs to provide a combination of excellent performance and low cost. You can take PanFS from your dock to serving data in a day with a plug-and-play solution that is easy to install, manage and grow. With its modular architecture and building-block design, investment banks, trading firms and exchanges deploying ActiveStor Ultra can start small and scale linearly by adding metadata performance, or bandwidth and capacity, easing the strain on quantitative analyses backtesting and allowing improved modeling.
