All-flash arrays are storage for today
Once considered a fringe or special-purpose technology, the flash array storage market has grown rapidly, and acceptance of all-flash arrays (AFA) as mainstream primary storage devices has become the norm. In the past five years, AFAs have made the move from a niche product supporting specialized applications to the most dominant form of storage for primary applications.
This isn't a minor thing: According to IDC, the market for AFAs grew to $1.4 billion in the first quarter of 2017, up 75 percent from the previous year. The market for hybrid arrays—arrays that combine flash with hard disk drives—was at $2 billion for the same quarter.
While the main drawback to flash is high cost, that clearly hasn't stopped enterprises from adopting all-flash arrays. This is helped by the fact that flash prices have been rapidly dropping while flash storage density has been increasing, with more vendors moving to technologies that can store three or more bits per cell.
At the present price, the future of AFAs looks bright.
That doesn't mean cost isn't a consideration. According to IDC, raw flash currently costs 43 cents per gigabyte while 15,000 rpm enterprise-class disks cost 23 cents per gigabyte. A 7,500 rpm drive costs even less. However, AFA manufacturers compensate for the cost by leveraging data reduction technologies like compression and deduplication on the fly, giving flash much higher effective capacity than hard disks while keeping high performance. As a result, the effective cost of flash approaches that of conventional hard disk drive arrays and, according to some vendors, is actually less in terms of total cost of ownership. But cost comparisons of flash vs. hard disks are complex and many companies remain skeptical. What is clear is that the market continues to grow rapidly as companies see the benefits of flash versus the perceived expense.
Six considerations for moving to an all-flash data center
As the all-flash array market has expanded, it has also differentiated itself, as vendors look for the most compelling value adds. The differences they highlight are not so much capacity and speed, because those are table stakes for flash. Some arrays are designed to be as easy to manage as possible, with most management tasks being taken care of automatically. Other arrays emphasize flexibility, giving the owner the maximum amount of control and the ability to use the arrays in almost any kind of configuration and environment. In general, AFAs will suit most common workloads, such as databases or virtual desktop architecture, and most major vendors offer several lines of flash products optimized for different levels of performance and capacity.
The result has been an entire ecology of AFAs targeting customer requirements, with a broad selection of capacity and capability.
Hybrid arrays, which consist of about 10 percent flash and 90 percent hard drive capacity, are currently the main competition for AFAs as primary storage. This hybrid-flash market currently outsells AFAs by offering near-flash speed at a lower cost. But it's the "near" that makes AFAs the storage solution of choice when speed is paramount. And the AFA market is growing more rapidly than the hybrid market.
Why flash arrays?
The nature of flash of any sort is quite different from that of hard disk storage. This results in a laundry list of advantages for flash and a few disadvantages.
The fundamental difference is that flash is all solid state. There are no moving parts in an all-flash array. Flash also doesn’t store data contiguously. Instead, it spreads it out over the array in response to sophisticated algorithms that minimize wear on the cells. This is because flash cells wear out after a certain number of reads and writes. There’s no advantage to writing data back to the same location; in fact, it’s a disadvantage, since it wears the cell. This makes no difference to the read or write speed, since flash is a random-access device. That's in contrast to disks, which have improved performance when they can read or write sequentially.
[ Keep up with hot topics on Storage with enterprise.nxt’s newsletter. Subscribe now ]
Reliability was a concern for early flash cells because of the wear factor, which was addressed by the manufacturers overprovisioning—providing extra cells—and improving chip design. As a result, most enterprise-grade AFAs are rated for a life span of five or more years in normal use and concerns about chips wearing out are largely moot.
That leaves the issue of cost. A flash chip is a complex structure, and it costs a lot to produce. The cost is following the classic semiconductor pattern of dropping rapidly as volumes ramp up and manufacturers get more experience. Still, the raw cost of flash is more expensive than hard disk drives. This is the major sticking point in faster adoption of AFAs, because while the cost of flash arrays is dropping sharply, the cost of disk drives is going down as well. The price drop has slowed temporarily because of a shortage of supply as demand grows and manufacturers retool for new versions of flash. Both technologies are cheaper than they were three years ago, and the differential is shrinking.
When replacing hard drive arrays with AFAs, data center users and operators see additional benefits in terms of size, weight, and cooling.
Flash arrays run cooler than disks, reducing cooling costs. Flash arrays are also lighter and more compact, saving floor and rack space. Some users have reported 75 percent or more reductions in rack space by switching to AFAs, and when using a colocation data center, that has a direct and immediate impact on facilities costs.
Weight is another concern. A fully populated drawer of flash drives weighs about 70 pounds while the equivalent capacity in hard disks weighs 2,700 pounds. That’s a big difference when you’re horsing around the racks, building modular or containerized data centers, or creating smaller, edge-focused "data center in a box" solutions.
Storage management is considerably eased as well. While disk arrays require constant management to achieve optimal performance, AFAs have so much performance that performance-tuning is simple. Also, many AFAs are built with modern management frameworks, which vastly simplify setup and day-to-day storage management.
Today and tomorrow
Flash started out in the enterprise as a niche product because of the high costs. Even at prices a few years ago, flash could make economic sense as storage for high-value, high-stress applications such as mission-critical databases. But as prices dropped and data reduction became ubiquitous, storage administrators got more familiar with flash and its role in the data center expanded to speed up critical applications. Some enterprises dedicate entire flash arrays to one application, but the changing economics and the need for speed has expanded the flash market out of application silos and into the general storage pool.
In general, AFAs are used for the majority of tier-zero and tier-one storage, supported by hybrid flash and disk for the rest. According to an IT Pro Research study, sponsored by Hewlett Packard Enterprise, 61 percent of companies surveyed have already deployed all-flash at some level and are increasing its use.
Hard disks are not going to go away, any more than tape went away. But like tape, hard disks will become second- or third-tier storage for jobs like archiving. AFAs, in different sizes, shapes, and capabilities, will dominate primary storage in the next few years.
All flash arrays: Lessons for leaders
- The economics of using AFAs are rapidly changing.
- There are a number of benefits beyond the speed of the storage for data centers.
- Hard disks will be around for the foreseeable future, but increasingly AFAs will become the dominant primary storage.
Gavin Cohen was interviewed for this article. He is the former VP of Product and Solutions, Nimble Storage, at Hewlett Packard Enterprise.
This article/content was written by the individual writer identified and does not necessarily reflect the view of Hewlett Packard Enterprise Company.