SSD RAID array: What it is and how to use it

RAID has been a fundamental of storage since its introduction in the late 1980s. But now, as solid-state flash storage becomes commonplace, questions arise over the use of SSD RAID.

In this podcast, SearchStorage.co.UK Assistant Site Editor Fran Sales interviews SearchStorage.co.UK Bureau Chief Antony Adshead about using an SSD RAID array, mixing SSD and spinning disk in a RAID group, and RAID for server-side flash.

SearchStorage.co.UK: Can you use RAID with SSDs?

Adshead: The one-word answer is yes, but of course there are numerous caveats to add and details to drill down into.

So, let’s start by recapping on RAID in general and reminding ourselves that it’s a method of enhancing data protection and/or performance by mirroring, striping and adding parity to data to drives in a storage array.  

And to recap on SSD, we can remind ourselves that solid-state drives are flash-based memory modules that provide rapid access to data. SSD’s value lies primarily in its high IOPS rates due to having no moving parts, with much better IOPS per pound or dollar than spinning disk. On the downside, they are much more costly per unit than spinning disk and so lose out to it on cost per gigabyte.

SSD comes in various formats: as drives that can be slotted into an array, as part of a discrete all-SSD array appliance and as server-side flash in PCIe card format.

It’s potentially straightforward to use RAID techniques with solid-state drives in an array. To a RAID controller they’re drives just like any other. The question is, however, would you want to?

Well, you’re less likely to want to make use of RAID’s performance-enhancing features with SSD, as solid-state drives already provide something like 80 to 100 times the I/O of spinning disk, depending on the usage profile. But you could group a number of solid-state drives together and gain further I/O speed advantage by striping data across them. There are also methods of mixing solid-state and spinning disk to make use of the best characteristics of both, but we’ll come onto that in a while.

But for now, dealing with data protection, if you’re considering running a tier of SSD for high performance, you’ll definitely want to consider RAID-protecting them, and that can be done via mirroring (RAID 1) or using parity protection, with RAID levels 5 or 6. Given the cost of solid-state drives compared to spinning disk, the parity levels are preferable, especially given that SSD’s high I/O rates will offset the write penalty inherent in parity RAID.

SearchStorage.co.UK: Can you mix solid-state with spinning disk drives [in a RAID configuration]?

Adshead: In theory you can mix hard drives and SSDs with a standard controller, but there’d be no real advantage to that and of course there’d be the usual RAID restrictions on drive size, with all drives only having the working capacity of the smallest one among them.

There are, however, some products available that let you mix SSD and spinning disk and use each to their best advantage: spinning disk for capacity and its suitability for writes, solid-state for its speed of reads.

Adaptec, for example, has software it calls hybrid RAID that creates a RAID 1 or 10 mirrored array, where writes go to the SSD and the hard drive but all reads come off the SSD. In tests, results have been mixed, however, with throughput figures that matched those of SSD according to whether they were read or write, but I/O rates falling short of those of pure SSD in all cases except Web serving.

Disadvantages of such hybrid RAID include that capacity will be limited by the capacity of the smallest drive, almost certainly the SSD, and also that if the SSD side of the mirror fails, your performance levels will drop off a cliff.

SearchStorage.co.UK: What about RAID with server-side flash?

Adshead: Server-side flash is an increasingly common application for SSD [that] puts it right next to the processor and takes advantage of its bus speed and often integrates with motherboard memory. Usually PCIe flash either has RAID built in for data protection; allows software RAID from the server CPU; or, in some cases, such as with Fusion-io, has a separate RAID parity chip on the card. And, of course, with multiple cards you can RAID-protect them as you would more than one drive on a server using software RAID from the OS.

Looking to the future, there is a definite weakness that needs addressing, and that’s having one flash card that is essentially direct-attached storage in a server that could fail. So, the next frontier here is going to be aggregating server-side flash across multiple servers. And while there’s no way of doing that with RAID right now as far as I know, Dell has made suggestions about linking many server-based flash cards together using the server aggregation technology it got last year in its acquisition of RNA.