RNA Networks Primes Memory Cloud

By providing access to more memory resources via its distributed memory service, the "memory cloud" increases the performance of the symmetric multiprocessing (SMP) architecture, explains Rob Butters, chief marketing officer and vice president, products, at RNA Networks.

Currently, when applications run out of memory, performance takes a hit, explains Butters, adding that memory limitations also place a practical limit on the size of models and scope of analytics that financial firms can run. By connecting to a memory cloud, many of these limitations are lifted. "Instead of firms analyzing portfolios in multiple tranches, they would have the capability to analyze individual portfolio components," says Butters. "It would also allow them to do it more quickly, such as doing portfolio and risk analysis on an intra-day basis."

In addition to improving application performance, Butters says he sees the memory cloud improving overall utilization of datacenter resources by moving data off hard disk drives and onto the memory cloud, thus eliminating the input/output (I/O) bottleneck that crops up between processing and storage media. "The issue is that applications are feeding data in terms of microseconds, but the disk I/Os are measured in milliseconds," he says. By moving the data to the cloud, the CPU cores are better utilized and the storage systems become more responsive since they are not handling the offloaded data, he adds.

Firms can also see improved performance improvement by scaling their server investments horizontally to add more memory to the cloud rather than scaling upward by investing in servers that have more memory capacity. "The pooled memory can come from different types of servers running different operating systems," says Butters. Firms can easily build memory servers from commodity servers to add to the cloud, he adds. "They can stitch together standard servers without needing to buy the larger ones," he adds.

RNA officials claim that the platform can scale to hundreds of servers and support "a fairly good portion of the datacenter real estate," says Butters.

MVX is a datacenter-focused platform, where firms typically connect the participating physical servers via a 10 Gbps cross-connect, according to vendor officials. "Using 1 Gbps connections just doesn't make economic sense," says Butters. Eventually, the vendor would like to scale the offering into a campus-oriented platform, but that is a ways off, says Butters.

The vendor has been speaking to a number of financial services firms, including a matching engine group and exchange operator, according to officials, who decline to identify their potential clients.

Rob Daly