The Flash Storage
As prices for solid-state storage drop, flash becomes
a valuable solution for data centers to maximize efficiency
by | Dan Tynan
Dan Tynan is a freelance writer based in San Francisco. He has won numerous journalism awards and his work has appeared in more than 70 publications, several of them not yet dead.
Many IT professionals once considered flash storage a luxury reserved for only the most demanding workloads, but flash has come a long way, and the perception of it is changing.
In recent years, solid-state drives (SSDs) have quadrupled in capacity while dropping steadily in price. Along the way, they’ve become a necessity in the data center.
When SSDs first appeared in the 1990s, they were much faster and significantly more expensive than their magnetic storage competitors. In the decades since, flash has typically been used to provide durable storage for mobile devices or in performance-intensive applications such as financial trading, where increasing the speed of transactions can have a dramatic effect on the bottom line.
“One of the places SSD caught on in a big way was with Visa and its automated transaction process,” says Jim Handy, a semiconductor market analyst with Objective Analysis. “Adding SSD allowed them to process 50 to 100 percent more transactions in the same amount of time. It was easy to measure the revenue gained versus the cost of the storage.”
But as SSD technology steadily improved over time, and manufacturing volumes ramped up, the cost differential between flash and hard-disk drives (HDD) began to shrink. In 2016, that gap pretty much disappeared, says Dan Cobb, vice president of media strategy for Dell EMC.
Even though flash storage devices have always had a number of advantages over mechanical drives — such as faster performance and lower energy consumption — the raw acquisition costs were always higher, he says.
Last year that began to change.
As a result, flash has become the storage medium of choice to consolidate general-purpose applications such as
e-commerce databases, risk management for credit card transactions and real-time security analytics, says Cobb.
“Now, in nearly every single metric, enterprise flash has an advantage over hard drives,” he says. “Customers are no longer asking, why flash? Instead, they’re asking, why not flash?”
Key Storage Performance Indicators: Flash vs. Hard Drive
Capacity: At its most basic, capacity simply means how much data a particular storage medium can hold. Traditionally, mechanical drives have had an edge over solid state, though SSDs — using techniques such as compression and deduplication — have closed the gap and made direct comparisons trickier. In general, hard drives can still store more data for less money, which makes them a better choice for cost-sensitive applications such as backups and archiving.
Latency: This refers to how much time it takes to complete a storage operation. Flash has a huge advantage over hard drives here, especially when using faster interfaces such as NVMe.
Throughput: How much data can a system process at the same time? Throughput matters more when moving large chunks of data (such as backups), less when doing random access operations (such as database transactions).
Input/Output Operations Per Second (IOPS): This factor measures how many read or write operations a storage device can handle over time. Even the fastest hard drive can’t touch solid state when it comes to IOPS. But the methods for measuring IOPS can vary wildly, making apples-to-apples comparisons difficult.
The Most Flash for Your Cash
With flash, getting things done faster is the name of the game, says Greg Schulz, founder and senior advisory analyst at StorageIO.
“When companies are buying flash, they’re really buying productivity,” he says. “When people ask me how much solid-state storage they should get, I tell them ‘As much as your budget will allow.’ ”
Despite its drop in cost in recent years, flash won’t entirely replace mechanical discs in many organizations. Even major cloud service providers are likely to use mechanical discs and possibly tape drives for less active data, such as bulk storage, archives and backups, says Schulz.
Data centers that can’t yet afford to go all in on SSD for more active data may opt for hybrid arrays that combine mechanical discs with a solid-state cache, offering a compromise between cost and performance, says Schulz.
But over time, hybrid systems will gradually be supplanted by all-flash arrays, notes Vito Ferrante, CDW’s EMC team lead and a flash storage technology evangelist. As with virtualization before it, going with flash is rapidly becoming an easy decision for data center managers, he says.
“People used to ask, ‘Should we virtualize this workload?’ Now they say ‘We’re virtualizing everything and you need to make an argument not to,’ ” Ferrante says. “Later this year and next, I think arrays will be all flash, unless you can make a good argument to use a mechanical disc instead.”
Among the workloads for which flash storage provides clear advantages are Internet of Things applications and cloud services. IoT systems generate massive amounts of data that must be stored and analyzed. The speed and efficiency of flash enables many advanced features and capabilities of IoT, such as capturing and analyzing data in real time.
Many organizations are migrating mission-critical applications to private clouds. When these workloads demand greater performance than HDDs can deliver, flash becomes an excellent option.
Ferrante adds that flash is also making strong inroads with hyperconverged appliances — integrated boxes that combine hardware, software, networking and power and that can be easily plugged into a rack, simplifying data center management.
“If I can walk a customer through a data center and say ‘I’m going to remove this piece of hardware and that network switch — and by the way, we’re going to increase performance 300 percent’ — I look like a hero,” he says.
The real challenge for data center operators then becomes how to get the biggest bang from their SSD investments.
“How do you know how much solid state you need, or where to put it?” asks Schulz. “You need discovery tools that can look at your environment and say, here is a bottleneck; here is where a little bit of flash can address a problem. And if you use a little more, you can get an even bigger bang.”
Still, if you’re an IT manager, these are good problems to have.
SSD vs. HDD: Which Is Cheaper?
Prices for SSDs have been dropping steadily, but so have prices for mechanical discs. Which is cheaper overall? It’s not an easy question to answer. The traditional measure of storage costs — dollars per gigabyte — is affected by a wide range of other factors, including the type of storage device, its relative speed and the controller hardware. And thanks to features such as deduplication and compression, the raw capacity of an SSD isn’t as important as its effective capacity.
But StorageIO’s Greg Schulz warns that comparing cost per gigabyte is the wrong approach when considering storage devices. Organizations need to look at the productivity gains SSD can bring, especially for highly active data where delays of even a few milliseconds can result in lost revenue.
“The worst thing you can do is compare SSD to hard drives on a cost-for-capacity basis,” says Schulz. “SSD usually loses to hard drives on capacity, but it always wins on cost per activity.”
Faster and Bigger
Going forward, SSD manufacturers will turn their focus to performance and capacity, deploying faster interfaces and high-density manufacturing techniques.
“The first phase of flash was about performance-intensive scenarios,” explains Cobb. “The second phase was making it more affordable. Now, we’re moving into a period of more optimization, thanks to new technologies. In 2017, we’ll be hitting the accelerator again on performance and business benefit.”
One of these new technologies, nonvolatile memory express (NVMe), allows for a high-speed interface that can shuttle data up to six times faster than older Serial ATA (SATA) and Serial Attached SCSI (SAS) buses. Because it’s so fast and versatile, the NVMe protocol will be used for everything from improving throughput inside servers to connecting external storage systems, says Schulz.
“NVMe allows more input/output, more bandwidth and lower latency,” says Schulz. “It uses less CPU, which allows you to be more productive on the back end, internally, externally or even over networks. NVMe is in your future; the questions are when, where, why and how.”
Flash chip makers are also moving full steam ahead with 3D NAND, which enables them to stack memory cells in successive layers on the same chip, greatly increasing their density and expanding capacity. For example, Samsung has produced a 15TB SSD, and Seagate has announced it is building a 60TB flash drive.
Intel and Micron’s 3D XPoint technology promises even more dramatic speed increases further down the road. Announced two years ago, XPoint is a still somewhat mysterious type of nonvolatile memory that may be 10 times faster and more durable than NAND chips, and could eventually replace DRAM. The first XPoint-based products are expected to arrive later this year.
“3D XPoint is the next big major step, just like NAND was,” says Schulz. “From the IT standpoint, it’s on the roadmap. IT environments will start seeing it trickling in soon.”
Meanwhile, Samsung has countered XPoint with Z-NAND, an extension of flash technology that the company says can double its performance, while enhancing durability and decreasing power consumption.
“2017 is a wonderful time to be in the enterprise IT space,” says Cobb. “Not only do we have NVMe making flash faster, we’ll have Z-NAND and 3D XPoint, making it faster still. That gives customers a lot of choice about where to put their most business-critical, performance-intensive applications.”
Call us at 800.800.4239 to set up a consultation with a data center expert.