Micro Focus is now part of OpenText. Learn more >

You are here

You are here

Is all-flash storage right for your data center applications?

public://pictures/ldp.jpg
Linda Dailey Paulson Freelance writer
 

Mention the word “flash” in casual conversation, and most people instantly think of the USB device they insert into a smartphone or their MP3 music player. But in the data center, flash-based storage is increasingly replacing traditional media—namely, hard disk drives—for key applications. These are configured in large arrays designed to replace traditional enterprise storage media.

Where flash storage arrays once augmented traditional storage as a caching complement to hard disk drives, the technology is now capable of cost-effectively hosting entire data sets for some applications. Flash is faster and more reliable than spinning media, and prices have plummeted, even as enterprise data storage needs have continued to rise. Many IT professionals are now considering moving data centers toward all-flash infrastructures. It's time to take a serious look.

What is flash memory?

Flash is a type of nonvolatile semiconductor memory. The first flash memory devices were produced in the 1980s, first by Toshiba and shortly thereafter by Intel. These nonvolatile storage devices were based on electrically erasing programmable read only memory (EEPROM), technology that allowed data to be written and erased electronically, as opposed to the ultraviolet irradiation methods used at the time.

Flash relies on a trapped charge on nonvolitile memory chips to store data, so it has none of the moving parts that characterize the spinning magnetic platters in hard disk drives. With volatile memory chips, stored information disappears when power is removed. With flash, the data is retained. Flash earned its name due to the relative speed with which it can erase data.

Flash is touted for its fast performance, a reasonable total cost of ownership that's now on par with conventional disk drives, consistent performance, small footprint, and low latency.

Since its introduction, flash memory capacity has doubled every year, according to Hai Li and Yiran Chen, writing in Nonvolatile Memory Design: Magnetic, Resistive, and Phase Change. Since it first came on the scene, they estimate, the cost per bit for flash dropped by a factor of 200,000.

That trend has led enterprises to consider flash as a complete replacement for hard drive arrays in some cases, rather than as caching complement to boost performance of spinning disks for a few high-performance applications. Rather than merely augmenting hard disk drives, flash arrays are increasingly replacing them, most often for primary storage.

Operational advantages for the enterprise

The primary benefit of flash is its elimination of latency-creating delays associated with spinning disk, such as the need to spin up the disk platters or execute what are called seek-time functions. Seek time is the length of time it takes the drive controller to locate and retrieve data from its position on a spinning disk. Users laud the higher, consistent performance capabilities of flash, which is why it's been used as a way to speed up database operations, says Vish Mulchand, senior director of product management and marketing at HPE Storage.

He says that one company he worked with moved to flash to accelerate its business analytics process. In addition to being able “to churn through data and make decisions five or six times faster,” Mulchand says, the organization saw its backup window reduced by 90 percent while increasing staff productivity by 50 percent. Because flash storage is easier to use and manage, the business was able to eliminate the need for database tuning and other complex optimizations that required in-house expertise. 

Flash versus disk: Comparing oranges to oranges

Comparing storage costs between flash versus disk-based storage systems can be a difficult exercise. For starters, you need to use the right metrics. In addition to comparing capacity costs, for example, you may want to evaluate the relative performance costs based on the number of input output operations per second (IOPS).

The current price for a 15,000 rpm, serial-attached storage (15K SAS) hard disk drive ranges from $1.50 to $3 per gigabyte, with software and support factored in, versus about $1.50 per gigabyte for flash, says Mulchand. Pricing for a slower, 10K SAS drive ranges from $1.40 to $1.50 per gigabyte. 

From a baseline of raw capacity, users should undertake an oranges-to-oranges comparison, says Chris Evans, writing in ComputerWeekly. It doesn’t make sense to compare SATA, for example, with flash. Relative capacity and cost per gigabyte may be a good starting point for an evaluation, but he suggests choosing either performance or capacity of like media—flash to flash, for example—before running calculations.

The total cost of ownership for flash, measured over four years, is the lowest of all storage media and is expected to continue to decrease through 2020, according to Wikibon CTO and co-founder David Floyer. He projects that the cost per terabyte will fall from roughly $151 in 2016—a metric that includes power, cooling, maintenance, and other costs—to $9 by 2020. What is driving the cost down, analysts say, are the economies of scale associated with rising consumer demand for flash chips, as well as new storage architectures that allow data sharing without performance degradation and data sharing across the enterprise, as opposed to using dedicated storage pools for specific applications. The total savings ultimately depends on the storage need.

Flash will not usurp all existing storage devices in data centers. Flash storage arrays currently make the most sense for replacing some hard disk drives used for primary storage. They can be used in concert with spinning media. For secondary storage, says Mulchand, flash offers similar benefits when comparing against hard disks. But for archival storage, where performance is less of an issue, tape, magnetic storage, DVD, and other archival media remain the less expensive options.

Starting the transition to flash

If your organization has yet to determine whether all-flash storage systems are an ideal addition to its data center, start by evaluating it in the context of a new or existing project that would benefit from flash's latency-reduction capabilities. All-flash rollouts make sense especially for applications that are time-sensitive and mission-critical, such as Mulchand's example above of the business that needed to accelerate its business analytics process.

Many organizations get started with greenfield projects, where it's much easier to implement an all-flash infrastructure from the ground up. IT organizations have also been adding flash incrementally, as a replacement for existing hard disk storage systems, in conjunction with their three-to-five-year tech refresh cycles.

Some enterprises are already making the transition to all-flash data storage. And as more organizations quantify and report the operational benefits and strategic business advantages of flash, gradual adoption beyond primary storage is likely to occur, particularly as prices continue to fall.

Have you considered all-flash infrastructure for your data center? If so, for what types of applications? I look forward to your comments.

Image credit: Flickr

Keep learning

Read more articles about: Enterprise ITData Centers