The advent of microprocessors and the client-server architecture in the 1980s brought about the end of the mainframe era of computing.
Mainframes are still in use around the world, but the availability of cheap, commodity processing power gave rise to open computing architectures. This meant businesses no longer needed to buy high cost, vertically integrated computing towers from the likes of IBM.
According to David Flynn, CEO of Flash storage memory device maker Fusion-io, the enterprise storage market has yet to be disrupted in this fashion.
“Storage vendors still sell vertically integrated, vendor locked, proprietary systems,” he says. “You pay ten times over the odds for a disk drive when you buy a storage array because it’s a proprietary, closed system.”
One effect of this, Flynn argues, is that the performance and capacity of storage systems are tightly coupled. When an organisation needs performance, i.e. a higher speed of input and output (i/o) from the storage array, it must buy more capacity, even if it has plenty of unused storage space.
“Let’s say you have an application that is running slow. Nine times out of ten, a server administrator will look at the server and see that there are CPUs sitting idle; it’s the i/o that is the bottleneck. So he says to the storage administrator, I need more i/o, and the storage guy says, give me a million dollars and we’ll buy another storage area network.
“The server guy says ‘I haven’t even used the storage capacity you gave me last time, why do we need to buy more?’ The reason is that storage infrastructure serves two masters, capacity and performance, and enterprise storage vendors don’t allow you to buy one without the other.”
Fusion-io, Flynn claims, decouples performance from capacity. Its Flash memory devices attach to the back of a server, and pull up a cache of the ‘active’ data it might need to use. This greatly accelerates the i/o of the server, and therefore the performance of the applications it supports.
Most of its 2,500-odd customers came to the company because a storage i/o intensive system, such as a transactional database or an unstructured search application, was suffering performance issues, Flynn says.
But the technology can be used to consolidate servers as well, he says, especially in virtualised environments. “If you can deliver faster i/o, then each virtual machine requires less memory to keep track of what it’s doing. That means you can increase the virtual machine density on the server,” Flynn claims.
Furthermore, by decoupling i/o performance from storage disk capacity, Fusion-io will allow enterprise organisations to adopt the ‘scale out’ model of IT architectural design, Flynn says. This means adding more commodity storage devices as the need arises, rather than upgrading to more expensive kit. Today, this model is most commonly associated with web companies –Facebook is Fusion-io’s flagship customer.
When Flynn talks about disrupting the enterprise storage status quo, his rhetoric is directed at one company in particular. “EMC is the bottleneck in the data centre,” he says.
That is not surprising. In February of this year, EMC launched a new Flash memory card system, named VFCache, that bears an obvious resemblance to Fusion-io’s technology.
Flynn says the launch of VFCache vindicates Fusion-io’s model of putting Flash memory on the back of the server, and not in the storage array itself as EMC had done back in 2008.
“They have admitted that our model is more adept at delivering data,” he says. “But they’ve rushed it to market, and they can’t even talk about it without making reference to Fusion-io.”
Flynn claims EMC’s VFCache is functionally inferior. But even if this is true, he still has a fight on his hands. The question is, can Fusion-io disrupt the enterprise storage market fast enough, or will EMC exploit its considerable installed base to beat the company at its own game?