When the topic is flash storage, “fast” is generally the first word that comes to an IT manager’s mind. The second? “Expensive.” While it’s true that flash memory costs significantly more than traditional spinning disk, IT decision-makers are starting to heed a warning that easily trumps these concerns: They should view flash not in terms of what they’ll spend to leverage it, but what they’ll lose if they don’t.
“There are several use cases for flash where, literally, time is money,” says James Bagley, senior analyst at SSG-NOW. These include transaction processing applications and real-time business analytics, “where the faster you can identify a trend and capitalize on it, the more competitive you are,” he says.
Flash is a type of nonvolatile memory used by solid-state drives (SSDs), a technology that, like a hard-disk drive (HDD), stores data persistently. The primary difference between the two media is that an SSD contains no moving parts, resulting in much faster read-write performance.
Enterprise data centers have relied for several decades on spinning disks such as HDDs, and for good reason. Over the years, HDDs have steadily increased in capacity and reliability while dropping in price. That’s why they’re still the physical media most commonly used to store data today.
But capacity is just part of the storage puzzle. Storage performance hasn’t kept pace with the vast leaps in computing power over the last several decades. One reason is that HDD rotation speeds can’t surpass 15,000 RPMs, while seek times had improved only incrementally. As the gap between computing and storage continued to widen, IT professionals’ frustration over application performance grew.
“That’s where flash came in, and it’s been a gift in the world of computing,” says Arun Taneja, consulting analyst for the Taneja Group. “It arrived just as the performance pressure cooker was ready to burst.”
More than 75 percent of IT organizations responding to a 2014 Enterprise Strategy Group survey said that, in deploying their first SSDs, the impetus was to improve the performance of business applications. They’ve been in mainstream enterprise use for several years, and ongoing technical advances continue to raise their profile.
“Our industry is arguably in its most dynamic period of change in the last 10, if not 20 years,” says Dan Leary, vice president of worldwide marketing for Nimble Storage.
“Flash is fundamentally a game-changer from a performance standpoint,” says Paul Feresten, senior product marketing manager at NetApp. “If you look at input-output operations per second, or IOPS, it’s common to see flash have 25 times more than spinning disk.”
Learn more about digital storage by exploring our latest infographic here.
Data center administrators leverage flash memory in three primary ways. They use hybrid storage arrays comprising HDDs and SSDs, all-flash arrays, and server-side flash.
IT departments’ need for high-performance and low-latency applications continues to grow. Dynamic and often-unpredictable data center workloads require storage media that delivers extremely fast response. Applications that require thousands of operations per second can’t rely solely on spinning disks.
These needs explain why spending on flash is spiking. The amount of flash now moving into enterprise storage is “breathtaking,” says Bagley. He predicts that organizations will consume $15 billion worth of flash in various incarnations by the end of 2014.
If flash is such a game-changer, why don’t organizations deploy flash arrays for every storage need? For one, SSDs cost roughly 10 times as much as HDDs. Then there’s the question of durability. Flash devices can handle only a limited number of program-erase cycles before they degrade to the point that they’re no longer usable.
As with other IT options, storage solutions have tradeoffs. While they may not be as durable as HDDs, flash drives handle demanding jobs quickly and efficiently. And they’ve proven capable of lasting several years in demanding IT environments, as evidenced by vendors that warranty their flash drives for five years. Further, the alternative for improving performance — throwing more HDDs at a problem — brings its own cost issues.
One thing isn’t up for debate: Both storage performance and capacity are of paramount importance to data center operations. The trick is to optimize both without breaking the bank.
When all-flash arrays hit the market, enterprises with mission-critical, performance-starved applications snapped them up. “In these cases, the IT staff’s pain was so great that they said, ‘Money’s no object. Just do something,’ ” says Taneja.
However, using these arrays for every application that could benefit from better performance — whether they need more IOPS, better throughput rates or lower latency — is the overkill approach to flash, and that comes with an overkill price, Taneja says. IT leaders need solutions that perform better than HDDs for critical transaction processing systems and latency-intolerant applications. But for many environments, hybrid storage arrays offer excellent performance at a lower cost.
“I’d argue that all-flash arrays aren’t necessary. Flash is still expensive, and for most workloads, it doesn’t make sense to have everything in flash,” says Steve Kaplan, vice president of channel and strategic sales at Nutanix. Nutanix offers its Virtual Computing Platform; flash and computing sit together on each node within a converged infrastructure, requiring no storage-area network.
“It doesn’t make any sense to take flash and stick it at the end of the network, where it’s subject to network hops and latency,” says Kaplan. “Flash should be kept with the spinning disk right next to the virtual compute.”
EMC offers both hybrid array options as well as an all-flash array through XtremeIO. “Our approach for hybrid arrays is to have a dedicated flash tier accommodate the fastest and highest-performance systems, and where customers just need lots of capacity, we have tiers with traditional drives,” says Itzik Reich, field chief technology officer with EMC’s Emerging Technologies Division. EMC’s storage tiering software intelligently moves hot data to the flash tier — where XtremeIO’s deduplication and inline compression functionality ratchet up speed — and shifts cold data to slower media.
Flash solutions must be flexible to accommodate IT’s numerous and sometimes competing priorities, says Gary Orenstein, senior vice president of marketing at Fusion-io. Between large hybrid and all-flash arrays is a wide spectrum of flash deployment options.
Organizations that want to boost the performance of existing HDD arrays can use flash read cache devices, which can be inserted into storage controller PCI Express (PCIe) slots for immediate improvements in IOPS, throughput and response time. Cache is a temporary solution that serves up data rather than storing it.
When combined with deduplication, flash cache helps organizations avoid boot storms, which usually occur at expected times of day, often associated with technologies such as virtual desktop infrastructure. “VDI is by itself a very demanding application, so if 500 virtual desktops in a call center are turned on at the beginning of a shift, that’s a boot storm,” says Taneja. An IT staff can leverage flash caching at problematic times so storage isn’t overwhelmed by the flood of input-output requests.
Storage experts consider flash cache to be both highly useful and flexible. Using three SSDs, IT administrators can create a pool of flash cache that sits in front of their HDDs. Data flows from the outside to the storage controller, which views it as a pool of available cache. When activity heats up, the controller engages the flash cache, and when it downshifts, data is placed in the HDD arrays. Read-only cache can reduce latency by at least a factor of 10, and is much faster than spinning disks.
Storage decisions should be app-driven, because different applications make very different demands on storage, says Leary. “If you’re running a real-time analytics database, you want flash. You might even want to deploy it right in the server, instead of going to an external storage array, to further reduce latency.”
Among the applications that could benefit from flash are:
“All these solutions are deployed on flash now,” says Orenstein. “If organizations are running these in their data center — and they all are — without using flash, they’re losing money, because not only are they going to be unproductive, they’ll get outsmarted by their competition.”
When XtremeIO became generally available in 2013, enterprises weren’t sure if they should use it for certain applications or every application, according to Reich. “In the end, the majority of our use cases were for databases and VDI, because that’s the low-hanging fruit.” But increasingly, he adds, flash is no longer just for these types of applications; customers are using the technology wherever it makes sense.
Flash chips appear in a variety of places, from storage controllers to PCIe cards, in every type of implementation, says Taneja. But as organizations take their use of this technology to the next level, some are experiencing confusion. “The biggest complaint I hear from IT organizations is that there are too many choices, and they want help figuring out what they need,” he says.
That’s where professional services teams can help. They can work with IT organizations to educate them on flash and overall storage dynamics, vendors and solution sets, ensuring that they make the right decisions for now and in the future.