Rolling out flash in the enterprise? It's a matter of application
Here's what flash viable looks like
Flash SSD is changing the way that we store data. Vendors would like you to use it for all of your applications – and you probably will at some point – but which applications does it make most sense for today, and which might you have to work a little harder on to justify its use?
Some rarified high-performance applications such as stock trading were particularly attractive from the early days of enterprise SSD when the equipment cost far more than it does today, says Jim Handy, general director at semiconductor market research firm Objective Analysis.
"People ran these sophisticated algorithms between market close and open. They could now run them several times a day," he recalls. "They'd buy the SSD and make back the cost in a few seconds."
Some high performance computing applications that relied on reading vast amounts of data quickly, such as genomics research, were also no-brainers for flash. And video editing – which used to rely heavily on spinning drives with large amounts of cache memory before flash became more readily available – also makes sense as a niche application for SSDs. Editors constantly scrubbing through frames of video need a system that can read data as quickly as possible.
As SSD technology evolved, more general online transaction processing became the sweet spot for enterprise SSD, because it maps to easily quantifiable performance improvements.
"In the enterprise, that's the reason why SSDs caught on – because there would be people doing OLTP. If I spend $7,000 on a Fusion IO drive, then this I could make this one server process more transactions," Handy says. Increasingly, though, flash is moving outside of these specific application areas to become more feasible across a broader array of computing workloads.
George Crump, founder and president of storage technology advisory firm Storage Switzerland, takes a broad view. He sees flash as the preferred platform for any active or near-active dataset, which in his world is anything accessed in the last month or two. That includes heavily used Sharepoint or Exchange implementations, too.
"Any time you're thinking of adding hard drives to an array to increase performance, you should go to flash," he concludes.
Part of the shift stems from the price of enterprise flash, which continues to fall. The time was when enterprise SSDs would be a good replacement for 15,000rpm drives. That moved to 10,000 RPM, and now we're seeing them replace 7,500rpm units.
"I don't think the applications will drive us towards flash. All applications. Everything is valid for flash," says Alex McDonald, director and vice chair at SNIA Europe. "I can't see one that doesn't benefit. The only thing we need to be concerned about is the cost per something." For some, that might be the cost per IOP, and for others it might be latency, or density. The specific needs of the application and the data centre environment will dictate that.
One infrastructure trend opening a door for enterprise flash is in hyperconverged systems, says Silvia Cosso, senior research analyst covering Western European storage at IDC. "As more workloads are moving to the cloud, cloud-friendly systems such as converged ones are increasingly taking ground both at the enterprise premise and at the service providers' side," she says. "In particular, it is well suited to service providers, where extremely low latency is absolutely essential."
You'll often find converged or hyperconverged equipment vendors touting their equipment – replete with flash storage – for virtual desktop integration (VDI). The key selling point here is the elimination of bootstorms – the IO-induced snarl-ups that can happen when a herd of employees try to boot up their server-based operating systems at the same time during the morning.
Show me the money
Enterprise SSDs may be getting cheaper, but there's still a price premium. When considering flash for an application, savvy IT departments will want to evaluate the return on investment for a flash array by examining the total cost of ownership.
Applications that have a consistent value per transaction are perhaps the easiest to quantify. The more of those transactions that you're able to complete by solving a storage bottleneck, the more the system is worth.
Frank Berry, CEO at data centre infrastructure analyst firm IT Brand Pulse, recalls working on a project at a theme park. "Their ticket reservation system was slowing down, meaning that they could book fewer tickets per minute, and that represents millions of dollars," he argues. "So they put Flash in there to maintain their reservation rate."
Transaction value isn't the only kind of justification for using flash in a particular application, though. Berry recalls a bank that ran analytics on its data each night to produce a report that had to be on a senior executive's desk at 8am the next day. Workload got to the point where traditional IOPS simply couldn't cut it. Instead, they switched to flash to solve the storage I/O problem, enabling them to finish in the early hours of the morning.
"In that case, they simply want to meet their service level," Berry says. It wasn't a cost-per-transaction decision. The IT department couldn't serve the business effectively without the beefier storage, so it had no choice but to buy.
The final kind of justification is operational, Berry says. A vice president might tell the IT department that it simply can't hire another storage administrator. Instead, it might have to find operational efficiencies elsewhere to avoid buying in the extra skills. Flash storage does away with many of the performance tuning and other administrative challenges associated with hard disk arrays, meaning that an IT department can save operational dollars there. This is a rarer justification, though, in his experience.