Flash Storage Explained

All hat and no cattle?
All flash and no pan?
All disk and no flash?
Read this article by 451 Research analysts Rhonda Ascierto and Tim Stammers to get the low-down on the hottest new storage technology.
Emphasis in red added by me.
Brian Wood, VP Marketing

Flash storage: The ‘Scarlet Speedster’ of disruptive datacenter technology

Flash memory, the storage technology that is widely used in smartphones and other consumer devices, is the emerging datacenter technology that suppliers and operators should be most focused on, in terms of assessing and planning for it.
Flash is likely to be widely adopted in the datacenter in the form of flash arrays, and, because of its low power use and low latency, it could prove widely disruptive when adopted at scale.
Flash storage – a faster, pricier rival to disk – is already forcing incumbent suppliers to change. New flash systems could eventually be used to store as much as 20% of all data, which could mean higher-density or possibly smaller datacenters.
Flash storage came out as the most potentially disruptive technology in the datacenter among those that were evaluated by Uptime Institute experts and 451 Research analysts based on three criteria:

  • Size of potential impact
  • How quickly the technology will have that impact
  • Likelihood that the technology will reach fruition and disrupt the market

The analysis was part of a project by 451 Research’s Datacenter Technologies team: ‘Disruptive Technologies in the Datacenter’. The report explores in detail our research and all of the technologies assessed. The selection criteria excluded technologies that were either already widely adopted but causing disruption (such as virtualization), or which were new but unlikely to disrupt datacenter design and economics, or the roadmaps of suppliers.
Technology and context
Spinning disk is the established technology for storing large amounts of data online. However, disk has become a performance bottleneck. This, coupled with the fact that the price of flash memory has plummeted during the past few years, has seen flash enter the datacenter as an adjunct to disk, used to store the hottest, most frequently demanded data.
Flash, a type of solid-stage storage, is much faster than disk: Vendors of hybrid disk-and-flash storage systems are claiming that by converting just 2-5% of their disk drives to flash, customers can increase the performance as much as fivefold. Recently developed, purpose-built all-flash arrays can be hundreds of times faster than disk.
Unlike disk, flash is still developing rapidly across the spectrum – from the flash drives themselves, through caching and other management software, to new all-flash systems. Better low-level data management is driving up performance and reliability, and allowing enterprise flash devices to switch to a lower-cost form of flash called multi-level call (MLC). Until recently, MLC was mostly used in consumer devices, but now it is being used to provide enterprise performance and reliability.
The real promise of flash is that it not only makes storage denser and faster, but could change the physical and energy footprint of the datacenter.
Entrenched storage system suppliers are being forced to adapt
Flash is already disrupting the storage industry; we estimate that more than half of midrange and high-end disk arrays today use some element of flash, and by 2018, the vast majority will (including new, purpose-built all-flash arrays). There are a number of ways flash is being used for datacenter storage, including:

  • Hybrid products that use both flash and disk in the same system. These include systems designed from the start to exploit the very different technical characteristics of flash compared with disk. However, the majority of hybrid arrays are conventional disk designs retrofitted with flash – and these currently represent the most common way of using flash in datacenters.
  • Existing disk arrays can also be configured to be powered entirely by flash, by replacing every disk drive with a flash drive. However, the results often do not justify their higher price tag. This is because these systems were not originally designed to use flash, and lack functions such as compression and de-duplication.
  • Purpose-built all-flash systems that offer significantly better performance and prices than disk arrays retrofitted with flash. These systems frequently include low-level features such as de-dupe and RAID that are designed to fully exploit the performance of flash, as well as controller algorithms that boost performance while also extending the life of flash drives. The results can be impressive: more than one million read I/Os per second (IOPS), which is faster and cheaper than the most expensive disk arrays.
  • Flash drives fitted within servers to provide even faster access to data. These often store a cache copy of ‘hot’ or frequently demanded data, for which the master copy is still held and heavily protected in a central storage area network.

Small startups have led the charge for these new flash products. Large, well-established storage system vendors are being forced to develop or acquire their own versions. In 2012, EMC paid a reported $430m to acquire XtremIO, a startup that developed a purpose-built all-flash system. In 2013, IBM said it would spend $1bn to boost its flash storage offerings. Almost all of the major storage suppliers have advanced all-flash systems on their product roadmaps.
New all-flash systems slash storage opex
After compute servers and cooling systems, disk storage is usually the next-largest consumer of power in a datacenter. Disk storage can consume as much as 30% of total datacenter power. Flash consumes significantly less power, cooling and space than disk. Measured per IOPS, flash storage uses roughly 600 times less power in kW than disk.
451 Research has profiled a deployment of a purpose-built all-flash array at a major manufacturer (unnamed) that needed to improve processing times and capacity of its large Oracle/SAP database implementation with high I/O demands. By swapping out its existing disk storage systems with an all-flash system from XtremIO (now owned by EMC), the manufacturer significantly reduced power and cooling costs and saved space in the datacenter. In addition to improving the responsiveness of its SAP/Oracle environment to improve processing and support new applications, the all-flash array showed the following results:

  • Positive ROI, saving the company more than $750,000 over three years
  • 82% less power consumed
  • 90%+ less rack space used
  • 82% less cooling capacity required

The greatest performance gains and potential opex savings are from purpose-built all-flash systems, although hybrid and other flash systems also consume less power and space than traditional all-disk arrays (just not to the same degree as new, advanced all-flash systems).
All-flash arrays and servers can also be run much hotter than traditional disks without performance or reliability problems. This opens up the prospect of datacenters without any cooling, since disks are the most likely components to fail, and heavily influence ASHRAE temperature guidelines.
Opex savings from flash would benefit enterprise and public-cloud datacenters. Reduced space, power and cooling requirements free up capacity. This can defer, reduce or eliminate the need for new capacity (in the form of equipment purchases, retrofits, IT rationalization, or even new datacenter construction).
Purpose-built all-flash systems could be particularly attractive to companies whose IT is housed in multi-tenant/colocation datacenters. This is because their providers typically charge according to space and power used, both of which could be reduced. These datacenter operators, therefore, could see a slowdown in revenue – although history suggests this space and power is quickly used.
How much data will be stored on flash?
The degree to which flash will replace disk – and therefore its disruptive impact – will be limited by flash’s relatively high price: a purpose-built all-flash array could cost 10 times more than a comparable disk array. Although the price of flash is dropping, so too are disk prices – and disk is currently falling faster than flash. For the next several years, disk is very much expected to be cheaper than flash on a per-GB basis, and therefore more suitable for storage of ‘hot’ data that does not require high performance.
However, a price crossover between flash drives and high-end, low-capacity disk (the fastest and most expensive type of disk) is approaching rapidly. Within the next five years, we project that flash will dominate the market for hot data storage.
How much data will be stored on all-flash arrays, and how much on disk? While flash today is being used to store the hottest of data (such as high-end online transaction processing, virtual desktop infrastructure, SQL databases and Exchange applications), we believe flash will increasingly be used to store all manner of hot, frequently accessed data.
Historically, hot data accounts for roughly 20% of all data stored. The fastest (most expensive) disk has been used to store hot data, while tiers of cheaper disk have been used for ‘cool’ data that is infrequently accessed. Because the bulk of data will always be cool, and disk will be cheaper than flash for many years yet, disk will continue to be used in datacenters – either in systems powered entirely by disk, or in hybrid disk-and-flash systems. Also, using flash inside servers means servers can reliably run above ASHRAE temperatures.
The 451 Take
Most CIOs will initially balk at the high cost of advanced flash storage products, yet adoption will grow once they match the price of high-end disk, which could be as early as 2018. But once energy and space savings are factored in, the benefits will become more apparent: Within the next five years, roughly 20% of storage in datacenters could be flash-based, which could significantly reduce cooling requirements in datacenters. While these savings could drive adoption beyond 20%, for the next several years, it is unlikely that flash will become cheaper than disk per unit of capacity, which will limit flash use to hot data to some degree.
There will likely be special use cases for all-flash datacenters, such as very hot or active-active datacenters. Many of the large, entrenched storage providers have announced plans to develop advanced all-flash arrays and server-side caching systems. Incumbent suppliers may face competitive challenges from smaller, specialist startups, but it is unclear that this will create a significant shift in market share, because the relatively slow adoption rate at present is giving the incumbents time to respond to market demand.