The needs of server and desktop virtualisation and their I/O demands have coincided with the decreasing cost of flash to bring solid-state storage into all parts of the data centre architecture. But solid-state is still costly and much data still must be kept on spinning disk.
So, the hybrid flash array has been born, in some cases the result of adding flash to existing technology, but also as a new product category in which hybrid flash arrays make use of cheap SATA drives and data deduplication technology.
In this interview, SearchStorage.co.UK bureau chief Antony Adshead speaks with Chris Evans, an independent consultant with Langton Blue, about defining the hybrid flash array, the chief use cases for them and key differences among hybrid flash array vendors' products.
SearchStorage.co.UK: What is a hybrid flash array?
Evans: [Firstly, there is the] traditional array, which consists of hard disk drives and [it's] the sort of thing that’s been around for the last 20 years since the Symmetrix was introduced by EMC, and everyone is quite familiar with that technology. That would have a mixture of drive types; [fast Fibre Channel drives, slower] SATA drives and so on.
At the top end we see all-flash arrays, where the device contains nothing but flash drives or flash memory, and that’s obviously a very expensive product, [being filled] flash.
In between we see the hybrid products, which haves some traditional hard drives and some flash, and the mixture of that depends on the usage of that device. So, a hybrid array is a mixture of both traditional hard drives and solid-state drives.
SearchStorage.co.UK: What are the key use cases for a hybrid flash array?
Evans: So, clearly, within the storage arena, pretty much everything comes down to cost versus performance; we’d love to reduce our cost, we'd love to improve our performance.
In a traditional array we can add flash and make it hybrid, and a lot of the major vendors have done that, added flash to them. That’s a great way of improving performance on a particular application or set of applications that you need to make run faster without necessarily re-architecting the design of that technology.
That’s a simple way of improving performance. Clearly there’s a cost trade-off there because if you spent money filling the whole array with flash devices, that would be extremely expensive. And typically in a shared storage environment, you find that I/O isn’t all demanding that high performance. You have a small subset of data that needs that performance and a majority that doesn’t need it, which is why hybrid arrays work quite well.
So, you need to bear one thing in mind with that style of hybrid flash array -- where additional flash added to an existing product -- and that is that you need some way of targeting the flash at the I/O that needs that high performance. We also see that vendors have brought out features like dynamic tiering, especially at the block level, which has allowed them to accelerate that.
Clearly the other key uses of this technology could be as a replacement of your existing technology. You might put in something brand new and you might go for a vendor with a specific piece of [hybrid technology, which] might meet your requirements for, say, a VDI application or some other process where you need to target I/O at a particular subset of your data.
SearchStorage.co.UK: What differences exist between hybrid flash array products?
Evans: As mentioned earlier, we’ve seen some vendors, especially the traditional and the established vendors take the approach of taking an existing product range and adding flash to that. So, people like EMC, Hitachi, IBM and so on [have] added flash drives to their existing arrays and that allows you to get higher performance out of that.
They’re interesting because they do give higher performance but they may not fully utilise the benefit of the flash. The reason for that is that a lot of those architectures were designed up to 20 years ago and were very focused around the fact that the hard drive was the slowest component in that architecture. So, things like the backplane, all the connectivity, the cache and the algorithms were directed to improving the fact that the hard drive was the slowest component.
So, you’ve now got flash in there, and those components possibly don’t perform as well as they could or can't drive the flash as well as they could. That’s one class of product; they work well and there’s nothing wrong with [them], but clearly they may not provide the best value [though] they may fit a requirement.
What we’re seeing is another range of products coming out from a new class of vendor and they're looking to bring out devices that are specifically architected to make use of flash. First of all, we’re seeing a software-based approach from companies like Nexenta, [which is using] the ZFS file system … to use flash and traditional drives. The flash is specifically targeted at being a write cache … to accelerate I/O though the array.
We’re also seeing other companies that use flash in a similar manner but as part of a bespoke hardware solution, [such as Tegile, which] has an array that does exactly that. It uses an architecture to accelerate the writes by targeting it purely on the flash.
So, we’re seeing a split in the market. The newer devices are using flash specifically within the box or the traditional vendors who use flash within their box simply to accelerate existing devices, and they’re very different in the way they work.