Violin Memory is looking to translate the record-breaking performance of its 3200 Flash Memory Arrays in IBM's General Parallel File System (GPFS) into solid sales growth. Using the arrays, IBM Research scanned 10 billion files 37 times faster than the previous record of 1 billion files in three hours that has stood since 2007. Breaking records is nothing new for Violin: In May, it set a world record dual-socket TPC-E benchmark with HP.
A high-performance, shared disk/storage file management system for multipetabytes of data, the GPFS offers fast and reliable access to a common set of files typically found in business intelligence, financial analytics, digital media, big data and seismic data processing applications. The test stand consisted of 10 IBM 3650 M2 servers, each with a 12-MByte processor cache and 32 GBytes of DRAM, and four Violin Memory 3205 solid-state storage systems with an aggregate raw capacity of 10 TBytes and aggregate bandwidth of 5 Gbps.
According to a just-published IBM whitepaper, while four disk drives would probably be sufficient to store the metadata for 10 billion files, they are not nearly fast enough to process it. Managing the metadata for 100 billion files would require about 2,000 disk drives and increase the space, power and number of controllers needed. Clearly, there is a need for different medium, solid-state devices that have the performance characteristics necessary to solve these problems.
This is a significant proof-point of flash memory in general, and flash memory arrays in particular, says Don Basile, CEO of Violin Memory. "Validation points like this from companies like IBM is a real big deal for us. They had the pick of the litter, and they could go anywhere."
Since announcing the 3000 series in May 2010 and starting volume production in July, Violin has been growing like gangbusters and is on pace to break the $100 million mark this year, he says. Violin has been working with IBM on this demonstration for more than a year, and now needs to evangelize the possibilities to target customers in markets like finance and government.
This is a significant development, says analyst Jim Handy, Objective Analysis. "The 37-times number is remarkable in and of itself, but the fact that this can be done at a much lower cost and in a significantly smaller space really says a lot. Most users of [high-performance computing] systems are constrained by budget, floor space and power. This could help those CIOs out a lot."
He says IBM is an important force in HPC, and this is an entryway for Violin to play into that space. "Typical HPC users will try to source from a single vendor so that a system failure will be addressed by one company without any buck passing. IBM's sourcing of Violin to the HPC community allows Violin to penetrate data centers that otherwise would have had nothing to do with the company."
See more on this topic by subscribing to Network Computing Pro Reports Parallel Memory Optimized (subscription required).