luchschen_shutter - Fotolia
The rise of flash in storage arrays is changing the characteristics buyers look for when evaluating storage. Solid-state drives (SSDs) are often purchased for different uses than hard disk drives, and demand different buying metrics.
Flash capacity plays second fiddle to the need for speed, quality of service and data reduction. Because of SSDs, flash storage performance expectations were raised for IOPS and throughput, and latency requirements have dropped from milliseconds to microseconds.
"One requirement I want is not just more capacity, but performance," said Albert Fuller, IT manager at the San Francisco-based California Public Utilities Commission (CPUC) and a SolidFire all-flash array customer. "I need more resources more than I need capacity. I look at IOPS. If I have a SQL Server, can I add IOPS on the fly, without shutting down the system? If I want to set up one virtual machine in production and run it, I need 500 IOPS for it to be functional. If I want to increase it to 1,000 IOPS, I need to do it on the fly."
George Crump, analyst with Storage Switzerland LLC, said flash storage adoption started as a point solution to solve specific performance problems. That was phase one of flash adoption. Now, it's evolved into a general-purpose storage medium and flash storage performance expectations have soared.
"We are now in the middle of the second phase of flash adoption: flash as a general-purpose storage system," Crump wrote in a Nov. 20 blog post on flash performance. "In this phase, all the flash arrays store all active data sets and workloads. We are also beginning phase three. In this phase, users are expecting flash performance. As a result, applications and environments are designed counting on flash storage performance. Since both phases one and two replaced hard-disk systems, the euphoria was easy to deliver."
Learn why Storage Switzerland's George Crumps says all-flash storage array purchasers will have to choose between performance and function.
Latency, throughput and IOPS have not only improved with flash, but they have become more nuanced. IT administrators need to take a different approach to understanding the metrics behind SSD and flash storage performance.
"People don't understand flash. They still think in terms of spindles," said Eric Carey, CIO for Valley Health System in Ridgewood, N.J., and a Violin Memory all-flash array customer. "The mathematics are completely different. It's not just plug-and-play. You need experience in flash or you won't get your money's worth."
Flash and SSDs often require a more thorough understanding of the application. For instance, the speed of an IOPS request is important when the data that is being requested is small in size. But when the application is large file-oriented, throughput is the key performance metric.
"When you are asking for a small file, then throughput is not the correct measurement," said Arun Taneja, founder of the analyst firm Taneja Group Inc., in Hopkinton, Mass. "For a 4K block, IOPS is more important. For a 128 KB, then throughput is more important. The one thing flash has done is speed everything up. IOPS and throughput have gone up. Latency is reduced. The biggest change has been latency, then IOPs, and then throughput."
The increase in performance also impacts the way applications are designed, Taneja said.
"Applications have to be much tighter to use flash more effectively," he said. "With flash, all expectations have gone up and it's starting to have repercussions on how we develop applications. The larger issue lurking is you don't design an application the way SAP was designed 20 years ago. Back then, you could develop an application and you knew, as a program, the answer came back in 50 milliseconds. With flash, you get a response in microseconds, so your program better be ready."
What's the key to optimum flash performance?
Developments in flash storage capacity and performance
Flash performance hindered by the I/O blender effect