Financial Services

Traders and risk managers in financial institutions run analytic models to derive and test investment strategies. The models may take days of compute time to fully analyze. When the useful life of a strategy is measured in months or even weeks, speed in running analytics impacts time to market. With teams of analysts running models at the same time, processing bottlenecks can be devastating. Accelerating and scaling analytics is of key importance.

MVX aggregates server memory into a sharable resource

RNA MVX directly improves analytics and model processing. MVX aggregates server RAM  into a shareable memory cloud that can be strategically applied to break processing bottlenecks. By keeping more market and trading data in working memory, performance is greatly enhanced. Historical data is available at near-memory speeds for market simulations and risk analysis. Data sets that may be needed by many servers at once can be effectively shared, removing the storage system as an issue during analysis.

All MVX uses can be applied to financial analysis.

• Memory Cache capabilities allow NAS-resident files to be accessed by any system in the data center at near-RAM speeds. There is no limit on file size; when given the RAM, MVX can serve data sets of 1 terabyte or more in memory.
• Memory Store capabilities provide servers with a “virtual RAMdisk” for storing and sharing intermediate results at blazingly fast speeds. A RAMdisk can be any size up to the entire Memory Cloud, so saved results can be far larger than any one machine could hold in memory at one time.
• Memory Motion allows any server operating system or database to attach a “virtual swap” drive that runs far faster than a physical swap drive. This permits a server to address a very large address space, which can speed up algorithms that need a lot of RAM.

Proven results

The financial industry has been an early adopter of RNA memory virtualization. Here are two examples.

A hedge fund has tens of terabytes of historical data that are used daily to evaluate the efficacy of their algorithms. Every day, 50 analysts use a 300-node cluster to run trading analytics on this database. They faced a replication condition, as multiple servers requested the same historical market data from storage, resulting in a slow process for all the analysts. The firm had implemented a caching solution based on local node storage, but quickly encountered problems scaling the solution. Distributing files among users and ensuring rapid accessibility due to high traffic between nodes when copying data led to persistent problems.

The hedge fund installed MVX and created a single memory layer accessible by all nodes. MVX replaced the in-house caching layer, holding a single copy of historical tick data, accessible to all users at memory speeds. Without duplication, the shared cache held the entire simulation data set in memory. The result was a drop in wall time by 20X, from days to hours, delivering models to traders in a fraction of the time. Taking only than three hours to install, MVX produced dramatic results with no changes to the client’s analytics engine.

In another financial services application, a shared job synchronization file was used to coordinate the processing behavior of clustered software over a large collection of stock price data. In this case, the job synchronization file was stored in the memory cloud which cut its access time to microseconds. At the same time, the price database was also cached in the memory cloud. This allowed repeated data requests to be served quickly from a shared memory buffer that was far larger than any single server, while also cutting accesses to the back-end storage system to a minimum.

When financial services firms analyze data, they need results fast. Conventional solutions suffer from limitations in memory size and in sharing methods that lead to bottlenecks and chronic delay. By aggregating RAM into a Memory Cloud in the compute tier, analytics and models complete more quickly. Without changes to data center hardware or modeling algorithms, analyses are done faster and revenue opportunities are exploited sooner.