In the search for increased financial application performance, systems architects are exploring a number of different hardware approaches. Parallel processing is one of them. Another is deploying Flash memory as a substitute for hard disk storage, which can be particularly effective where data throughput is a bottleneck. As it happens, quite a bit has been happening in the world of Flash memory of late, and recent business and technical developments are likely to impact the real world before too long.
Before we get into the latest goings on, it’s worth a quick recap of what Flash is, and where it stands in the hierarchy of storage. First off, Flash is solid state storage, based on silicon chips, which means it uses a lot less power than hard disk (rotating magnetic platters). Next, like hard disk, it is non volatile, so it retains data when powered off. It is also a lot faster than hard disk, though it’s not nearly as fast as (volatile) Random Access Memory (RAM), which is costly and limited in quantity.
Most importantly, to software applications, Flash “looks like” a hard disk, so there is no need to rewrite code to take advantage of the performance boost. That’s an important productivity and time to market attribute. It also makes it straightforward to boost performance of packaged applications, where code changes are not possible.
Flash memory has actually been around for some 30 years and it’s now ubiquitous as (sometimes removable) storage in consumer devices, such as music players and digital cameras, and even laptop computers (like Apple’s MacBook Air). Increasingly – as its capacity has increased and reliability has improved – it is also being used for enterprise-class servers, usually mounted on PCIe feature cards, or inside network-connected storage appliances. Nowadays, tens of terabytes of Flash storage are possible, albeit at a higher cost than hard disk. But you get what you pay for.
For financial systems, Flash has a wide variety of applications, with particular emphasis on those involving data management. Everything from pre-trade analytics on large time series, to portfolio and risk management, to counterparty information databases, to post-trade compliance and settlement will see a big performance improvement by using Flash for storage instead of hard disk.
Given its long history of development, one wonders how Flash can still be an innovative technology to track. But it seems that a lot is still happening in the space, and more is set to come. A few highlights are …
Diablo Technologies and SanDisk: Last year SanDisk – a major player in the consumer Flash market – licensed technology from Diablo Technologies that allows Flash to be inserted into RAM slots on a server motherboard and accessed via the main memory bus. The result is now shipping in servers from IBM and Supermicro, and means lower latency and less jitter for applications – important considerations for trading applications.
Fusion-io and SanDisk: Earlier this month, SanDisk continued its enterprise push by acquiring Fusion-io for $1.1 billion. Given Fusion-io is cited as a leader in the server Flash space, with revenues around half what was paid for it, it was a good deal for SanDisk, and probably reflects the intense competition among Flash vendors – others being IBM, Intel, Kaminario, LSI, Nimbus Data, Pure Storage and Violin Memory. And a good few more – it’s a crowded space. Under SanDisk, Fusion-io’s profitability should increase as it can now tap its parent’s economies of scale in memory chip manufacturing.
DSSD and EMC: Another recent acquisition was that of secretive Flash startup DSSD by storage giant EMC. It’s not that often that a startup is acquired while still in stealth mode, but when it’s been founded by Andy Bechtolsheim, whose previous startups were Sun Microsystems and Arista Networks, then there’s probably good reason to believe it will be a hit.
Details of the DSSD technology are scarce though pundits suggest that the company’s products – which could ship next year – will be external Flash appliances that mimic internal server memory, putting a new twist on in-memory databases, which can leverage the non-volatile and high capacity features of Flash versus RAM.
So expect to see all these recent developments lead to the availability of even lower latency, higher capacity non-volatile storage to power sweet spot applications that call for terabyte and petabyte big data processing and analytics in real time. And one can think of a few of those in the financial markets.
Pete Harris is Principal of Lighthouse Partners, an Austin, TX-based consulting company that helps innovative technology companies with their marketing endeavors. www.lighthouse-partners.com.