It will hardly have escaped anyone’s attention that EMC World is taking place this week in Las Vegas. Although I’m not attending this year I have been following the press releases. Probably the most interesting so far has been that around flash technology, something EMC announced at EMC World 2009 would have huge adoption in the industry. As we now know, for various reasons that adoption has been nowhere near as widespread as forecast (although EMC claim to have shipped 14PB of flash themselves).
In this image I took from Joe Tucci’s keynote speech at EMC World 2009, you can see the expected price erosion. Despite this, flash still hasn’t been widely used.
So what have EMC said this week? The flash announcement is more of a statement of direction than details on a specific product. EMC will bring MLC flash to their products alongside existing SLC drives. The main differences between these two technologies being cost, performance and reliability. MLC flash tends to get used in consumer devices and so it’s really not up to a 24/7 access profile. EMC are not the first company to do this however; Violin Memory have an MLC based product, the 3140 array.
EMC did make an announcement on a future product, currently codenamed Project Lightning, a PCIe server-based flash card. Many of you will have read (and be aware of) Fusion-IO from this post (http://www.thestoragearchitect.com/2010/10/28/storage-networking-world-europe-ii/). Again, EMC’s offering is not a first for the industry as Fusion-IO have been selling their flash cards (in a variety of formats) for some time.
What’s potentially different about Project Lighting is the integration into the server and virtualisation layer. I’ve only Chad’s rather speculative post to base my thoughts on, but the idea appears to be to use server-based flash in a tiered storage model to improve VMware performance. It helps to own the company of course, as you can direct the host developments towards matching the hardware developments; I can’t imagine many other array vendors have tried to run vSphere in the hardware, for example.
But where does Project Lightning take us to? Well, it reinforces the view that there’s a direct dependency between server virtualisation and storage. It means we’ll see hybrid storage/server solutions that make the best use of resources without having to deploy dedicated (and sometimes slightly siloed) technology towers. There’s also some thinking required around the way this kind of infrastructure is deployed. For example, in a traditional model of data replication, synchronous replication allowed the host data to be duplicated to another site. If the primary site is lost, the remote copy of data can be accessed and appears to be a “crash” copy, which should recover successfully. Virtualisation changed that and made it more difficult to replicate at the array level. What now happens when we’ve a dynamic infrastructure that replicates data between storage/cache in hardware and disk in the server? Where’s my data if any part of that infrastructure crashes? These issues aren’t unsolvable, but do require more thought in terms of architecture design.
Assuming EMC can bring Project Lightning to market quickly, I expect they will see competition from HP, who have already approached the server/storage combination with their P4800 storage blades, their partnership with Violin and their work on memristors. Whether we like it or not, the big vendors are moving us closer towards a converged world.