451 Research surveys hundreds of IT professionals in medium-sized and large organisations every year. And in the latest Wave 17 Storage Study more than half of those questioned said their organisations had deployed flash while a significant number said they plan to do so in the next 18 months.
As a result, although enterprise understanding of the cost, role and benefits of flash storage is still evolving, this reinforces the belief that the question for most IT managers is not whether they should deploy flash storage, but how they should deploy it.
Hybrid arrays rule
Datacentre operators are not short of options for how and where to use flash. The choices fall into three major categories: inside servers; in hybrid disk-and-flash storage arrays; and in all-flash arrays.
When enterprises were asked about their use of emerging storage hardware technologies, by far the most widely implemented were hybrid flash storage arrays, or arrays powered by a combination of disk and flash. More than one-third (37%) of respondents said they had implemented hybrid flash arrays and another 15% said they planned to do so within 18 months.
Server-side flash still growing
Second to hybrid arrays, the next most popular way of using flash in the datacentre is to install PCIe flash drives in servers. Also called server-side flash, this has been adopted by 18% of survey respondents.
The question for most IT managers is not whether they should deploy flash storage, but how they should deploy it
PCIe flash was the first major method of bringing flash into the datacentre, and as such it predated the use of hybrid storage arrays. However, server-side flash has since been overtaken by hybrid arrays, and the gap in adoption rates is set to widen.
The survey revealed that fewer respondents have long-term plans to implement server-side flash than hybrid arrays. However, spending by those that are already using server-side flash, or plan to do so, is rising sharply.
Pros and cons of flash options
Hybrid array and server-side flash use have relative advantages and disadvantages.
For hybrid arrays, the major advantage is their simplicity of implementation and operation, as well as the ability to increase performance for any application that is given shared access to the array.
In contrast, server-side flash only improves performance for applications running on the servers in which the flash drives have been fitted. However, that apparent disadvantage can also be an advantage, since it provides a means of precisely targeting specific applications – or virtual server physical hosts – that need a performance boost.
Multiple large and small storage suppliers are betting on this benefit by developing flash caching software that allows server-side flash to be deployed while making zero disruption to existing storage setups, and by allowing that software to cooperate with the caching mechanisms in standalone arrays.
In the medium to long term, these developments are very likely to accelerate the adoption of server-side flash. Currently, however, 451 Research data shows that two-thirds of enterprises have no plans to deploy it, compared with less than half of enterprises which say they have no plans to use hybrid arrays.
All-flash arrays arrive
A small minority (6%) of respondents to the study declared their organisations had deployed all-flash arrays.
Because they are powered entirely by flash, these devices are more expensive than hybrid arrays per gigabyte of capacity. However, compared with hybrid arrays, all-flash arrays provide consistently low latency, a performance characteristic that can be critical for certain applications, such as databases, virtual desktop infrastructure (VDI), server virtualisation and analytics.
Hybrid arrays attempt to ensure that as much hot or in-demand data as possible is stored in flash, but they must also store data on disk. As a result, there are occasions when even the best-designed hybrid arrays are forced to route data access to disk, with a consequently negative effect on latency. That does not happen with all-flash arrays, which store all data in flash.
As the per-gigabyte price of flash memory falls, and the I/O demands on storage systems rise, adoption of all-flash arrays should increase.
Outlook: 2014 make or break for flash startups
When flash first appeared in enterprise storage systems around five years ago it was already set to take on a major role in the datacentre. Because of the continuing drop in price of raw flash chips, as well as advances in product development, it is now fulfilling that expectation.
But for the foreseeable future flash will be more expensive than disk, for which the per-gigabyte price is on a similar downward trajectory. For this reason, flash will augment, but not displace disk in the datacentre for many years, and will mostly be used for a subset of performance-sensitive data.
More on flash arrays
The most popular way in which incumbent storage suppliers have embraced flash is by offering it as an option for use in their existing disk arrays, to create hybrid arrays. These first shipped from major suppliers in 2008.
More recently, a new breed of hybrid devices has emerged, which have been designed from scratch by a small handful of startups. Because of their ground-up designs, these suppliers claim that their products are more effective at marrying the very different storage media of flash and disk in a single system than the retrofitted disk-originated hybrid arrays sold by the incumbents.
Most notably, Nimble Storage, which was early to market with a ground-up designed hybrid array, has reached a revenue run rate of $100m, and recently filed to go public.
This presents solid evidence that startups are bringing something new to the party, and are gaining traction in one of the most competitive segments of the IT industry.
All-flash array adoption is still at the fringes, and this is still a fragmented space.
Another all-flash array pioneer – Violin Memory – now operates under the full scrutiny of Wall Street as a public company, and the pressure is on all of these players to start ramping up sales.
So 2014 is likely to be a make-or-break year for many of them, especially since the incumbent suppliers are now entering the all-flash array market in earnest with products that are either modified versions of existing disk arrays or have been newly designed.
While this may help the startups by validating the all-flash array concept and educating the market, it also presents formidable competition – for example, from EMC with XtremIO, NetApp with its promised FlashRay and even networking giant Cisco with its recent purchase of all-flash array maker Whiptail.
The next year is going to be very interesting indeed.