Cache Splash

It’s funny, I had a brief discussion about my blog with my IT director at work today; I wasn’t aware that he was aware that I blogged but it seems a couple of people outside of work had outed me, in what appears to be very complementary terms; he was pretty relaxed about my blog and one of his comments was that he assumed I discussed new products and I said I did.

But on the way home, I thought about it and to be quite frank, I used to talk a lot about new products but I don’t really do so these days. So it is ironic that today, I’m going to knock out a quick blog about EMC’s VFCache announcement; they don’t need the publicity but I’m going to talk about it anyway.

VFCache is very much a version 1.0 product from what I can see; EMC appeared to have set their bar quite low in what they are trying to achieve with this release; it appears that they’ve very much targeted Fusion-IO pretty much directly and decided to go after them from the get-go. Trash them early and don’t let another NetApp happen.

Go for engineering simplicity and don’t fill the product full of features….yet! Keeping it simple means that EMC can accelerate any array, not just an EMC array but in the future when new features come along, many of these might well only be available with an EMC back-end array. You’ve bought your Flash card, if you really want value….you need to partner it with an EMC array.

And in fact, to really leverage any server-side flash product; you probably do need array-awareness to ensure that you don’t do economically silly things like storing multiple copies of the same information in different caches; how many times do you want to cache the same data.

You need an efficient way of telling the array, ‘Oi I’ve cached this, you don’t need to’; this will allow you to utilise the array cache for workloads which might not easily support server-side caching currently. Perhaps at some point we’ll see a standard but standards are rarely fast moving in storage.

I also expect to see EMC build in some intelligence to allow it to leverage the split card capability; perhaps using PowerPath to flag that actually you might want to consider using the split card capability to gain performance?

I’d also be interested in seeing advancing modelling tools which allowed you to identify those servers and workloads which would most benefit from VFCache and what the impact is on the other workloads in the data-centre. If you accelerate one workload with VFCache and hence free up cache on the shared-array, do all workloads benefit? Can I target the deployment at key servers?

Deduplication is coming but it needs to be not at the expense of latency.

And of course there is the whole cluster-awareness and cache-consistency thing to sort out and perhaps this whole thing is a cul-de-sac whilst we move to flash-only-shared-storage-arrays…that’s until the next super-fast technology comes along.

Yes, EMC’s annoucement is very product 1.0 and a bit ‘ho-hum’ but the future is more interesting. Storage, Snorage? Sometimes but it’s impact sometimes wakes you up with a bit of a shudder’.

I wonder who is going to announce next or what the next announcement might be. 2012 might be a bit more interesting.