So IBM has a new mid-range array in the form of the Storwize v7000; the question is did it need one and what happens to the rest of the range? In the midrange space, IBM now have two of their own arrays in the form of XIV and the v7000; they also have the DS5k which is OEMed from LSI.
IBM beginning to disengage from their storage OEM partners was eminently predictable; IBM have finally woken up to the fact that they need to own the stack and partnering was not going to allow them to steer their course in the coming 'Stack Wars'. However, the DS5k probably does have a part to place at least in the short term in some niches; the EXP5060 high-density enclosure certainly has a part to play in this.
XIV, though, is looking vulnerable and not for the reasons that I think some people expect me to state. XIV, simply does not make sense for IBM economically; it costs too much to make I suspect. But 'Bod, XIV is made from commodity components; it should be as cheap as chips to make.
Yes, XIV is made from commodity components but it's made from an awful of them to build an array; the v7000 is also made from basically commodity components in a custom form-factor and it uses a lot less of them to deliver better performance and more capacity. The Bill of Materials for an XIV is going to be significantly higher than for the v7000; this makes v7000 more profitable to IBM and/or will allow them to discount the v7000 more substantially.
The latter is going to be hard with the current XIV set-up which pretty much runs with their own sales-force; IBM will find themselves competing with themselves but I am sure that they will be happy to accept the higher margins that the v7000 could deliver.
Of course, there are some savvy customers out there who are going to do the same sums as me and are going to ask some hard questions; if you can supply XIV for 'x', why do I have pay 'y' for v7000 when it costs you less to build and supply the v7000.
Now, I know there are some readers who are going to mention that XIV is rumoured to be delivering some much awaited improvements such as improved inter-node connectivity and clustering; I've also heard that something along the lines of NetApp's PAM may also be delivered. You could stick an XIV full of SAS drives which would improve performance but increase cost; you could build an XIV based around 2.5" drives which might have been an interesting proposition, improving performance and capacity. But none of these is going make XIV any cheaper to make and any more an attractive option for IBM.
XIV's killer feature is it's ease of use and self-tuning capability but are users willing to pay a premium for this to enable the range to continue to be economic for IBM to make? Could IBM actually position the XIV as the AS400 of storage; the box you stick in the corner and forget about? I guess, that's possible but storage is a phenomenally price competitive market despite the attempts of people to pretend it isn't.
And the v7000 does not look exactly hard to use and configure; I'm also sure that IBM have learnt lessons about self-tuning systems from the XIV which can be applied more economically to it's more conventionally architected storage systems. I expect IBM to keep XIV around for at least another generation; it's an insurance policy at least but I'm not sure it's got a long term future in it's current form.
XIV could find a new lease of life though as a combined compute/storage appliance; that might make a lot more sense. Perhaps as a grid-based deduplication device? Or it might simply go down in history as this storage generation's Iceberg? An array full of interesting ideas which never quite fulfilled it's potential and in ten years from now; we'll be say remember the XIV, it had that feature first.
But ultimately, it won't be the users who will kill the XIV as a storage array; it's IBM themselves, it just doesn't make sense.
I think you’re right that XIV is going to struggle to justify itself in the IBM line-up against the V7000 at a price and performance level, but it will still have a market for it amongst people who have looked at “normal” arrays like the V7000, DS5000, V-Max, etc, and decided that the cost of the various management tools and storage staff costs were too high to justify the increased performance.
The XIV can be treated largely like a single 161TB hard drive, it’s about as easy to configure and there’s no real performance tuning to be done on it. The V7000 is pretty easy to setup, but it’s got far more options for tweaking and squeezing the extra 10% of performance out of it.
To the finance directors of the world, the XIV will still appeal to them as it has done for the last couple of years, regardless of how much “technically” better the V7000 is.
This isn’t a technical argument; IBM can simply make more money out of the v7000. And it’s going to be hard to sell the XIV at enough of a margin to make it worthwhile for IBM; they could position it as the AS400 of the storage market but is that enough of a market?
The cost argument is of course highly speculative. XIV uses lots of Intel processors and distributed caches and lots of low cost SATA drives. Storwize V7000 uses only two processor modules, but to get similar performance to XIV you will need to use SAS drives rather than SATA. So there are swings and roundabouts in play.
For smaller customers, Storwize V7000 is the clear choice, and for bigger customers it’s XIV.
When there is a mid-sized situation that could be addressed by both Storwize V7000 and XIV, the personality of the customer will be the deciding factor.
The XIV success story has really been written by customers who are impatient with technology and they want us to all shut up with our fancy storage talk and just give them BIG SIMPLE NOW : )
It will be very interesting to see what the benchmarks say and especially when you start dealing with random I/O which is not coming from cache. Let’s face it; if it’s not in the cache; you are just talking about a big pile of slow SATA disk.
And of course the cost argument is speculative but I’d bet I’m not far wrong.
Was $350M a lot to pay to get a fancy GUI and a big kick towards usability? Actually, I’m not sure that long term it was….it changed IBM’s development priorities, moved them away from looking at fluff in belly-buttons back to looking at what the market wanted.
You are writing XIV’s epitaph way too soon… We are now up over 3,200 XIVs shipped and climbing. Why so keen to see the back of the biggest innovation in disk storage since Symmetrix came out 20 years ago?
The trick with XIV is to make sure I/Os almost always come from cache (lots of distributed CPU power = very aggressive distributed read-ahead) and while the occasional individual I/O might have a high latency, on a big multitasking system the average latency will stay very low.
The closest you will get to an XIV benchmark is probably my blog post from April http://storagebuddhist.wordpress.com/2010/04/06/xiv-ms-exch-2007-replicated-performance/ but being a non-trad box, XIV has a very non-trad performance profile. We recently heard from a customer who has got more than 100,000 IOPs out of their XIV in the course of normal business duties – something I had previously assumed was only possible in the labs.
I think you are talking about different customer needs.
1) XIV is being positioned in the High End market, despite most people tastes. The idea is about a self-healing system, no-brainer configuration and auto-everything in a per-case basis.
2) V7000 is a DS5K competitor/replacement, IMO.
Remember, what’s in a high-end system? Just performance?
There’re lots of cases where mid-range systems outperforms high-end parents, but with different capabilities.
XIV with 17K sustained IOPS vs DS5K is really the same, but that’s not the point.
3) XIV is wide-striping NO-RAID configuration. Think of mirror-dependent policies and forget about this machine. Try with lots of small servers and virtualized environments,replication with high-latency communications, storage administration nightmares and this is your product.
4) V7000 uses XIV-alike GUI, but it is NOT a cluster. SVC features (TP, Easy Tier, external device virtualization if you want to use and pay for it), storage migration and DS8K caching. Per-system licensing and NO MORE “per host” license!!!
Martin,
Did you intentionally title your post Wither XIV rather than Whither XIV?
Nigel
I meant exactly what I wrote, I normally do.