Storagebod Rotating Header Image

A Matter of Scale

IBM's recent announcements pretty much confirm a trend I've been seeing for sometime, Scale-Out-Storage is the new storage virtualisation; both in NAS and now in SAN. Pretty much of the major players have announced products in one of these spaces and in some cases both.

    1) EMC's V-MAX is a scale-out architecture for SAN

    2) NetApp's OnTap 8 brings scale-out architecture for SAN & NAS

    3) IBM's VDS brings scale-out architecture to SAN and SONAS is scale-out NAS

    4) HP have iBrix and Polyserve in the NAS space

And, I think pretty much everyone is expecting the new Hitachi box to be a scale-out version of USP-V which means that HDS, HP and Sun (assuming the relationship continues) all have a scale-out architecture for SAN. The other aspiring competitors, the 3PARs, Compellents etc all have ambitions in this space as well.

But Storagebod, IBM haven't announced a scale-out architecture to SAN or that's not what they appear to have announced. But they have, they've had a scale-out architecture in the form of SVC for some time and I think what they have announced in VDS is basically a new scale-out array. They won't admit this but there is no doubt in my mind that this is what they are in the process of doing. 

For some time, I have been suggesting that all the SVC is, is an 'array controller' built out of commodity hardware but with no disk. The perfidious Scot has never denied this but always just said to watch this space. 

So in the home of SVC, IBM announce a bundle of SVC and LSI disk and call it VDS; allegedly targetting the mid-range but theoretically, you could cluster these together and build a much bigger array if you wanted. Of course that would cause XIV some problems and this is why this is a territory specific launch; it's a toe in the water. 

IBM and NetApp?

The relationship between IBM and NetApp has been a profitable one for somebody. It allowed IBM to actually get some experience of selling NAS as opposed to giving NAS away; it also gave NetApp penetration into some blue-chip accounts in which they'd struggled to get a toe-hold in the past but do the recent announcements indicate a potential parting of the ways in the future. Also, I think IBM have declared themselves out of the running of any NetApp acquisition.

Firstly IBM have never been keen to push NetApp as a SAN device, that eats into their own product lines but if a customer wants it, they've been willing to sell it but they push XIV very aggressively in the space that'd you see NetApp as block devices. VDS, DS8k and XIV are enough SAN arrays for any vendor to try and sell.

Secondly, I've not felt much love towards OnTap 8 and NetApp's clustering from IBM. SONAS competes directly with NetApp in this space now; OnTap 8 is new and so is SONAS. Although both products are built on technology with some years of pedigree, I'd expect teething troubles with both. 

Thirdly, if you look at this quote from Steve Legge in The Register article,

   'For NAS there is the N Series up to around £250,000 and above that is SONAS'

I'd like to know if that £250,000 is list or discounted price; if it is list price, then IBM are going to be pushing SONAS into those accounts which were buying the equivalent of the FAS/V3000 series. This is NetApp heartland territory and would leave IBM only pushing the lower-end n-Series.

And lastly, is it not pretty much close to the fifth year anniversary of the NetApp/IBM relationship? 

Unified Storage Platform?

IBM could also do an EMC with the VDS and SONAS and build a Frankenstorage platform but at least they would have the advantage that both SVC and SONAS have Linux at their heart, as opposed to Flare and DART which run on different operating systems. I wonder if we will see a coming together of SVC and SONAS as an integrated product? That would taken them right into the heart of NetApp territory.

The interesting thing about building an array out of both SVC and SONAS is that long-term as a customer; it means migration and upgrades become a lot easier. Another one of the scale-out NAS vendors, Isilon have been pushing the building block approach and the ease of upgrade of their hardware; it's an attractive sales pitch to hard-pushed teams. IBM could do something very similar. Non-disruptive and automated migrations/upgrades of both block and file are both possibilities here.  


So are we beginning to see greenshoots in IBM's storage line? We might be but IBM's problem has never really been technology; it's just that their internal organisation seems to wield a can of DDT with regards to storage and is able to kill off any signs of life pretty damn quickly.

However, directly attacking one of their partner's core territory is pretty damn ruthless and that's not what we've seen from Blue for some time. 


  1. Chuck Hollis says:

    I’d agree with you that scale-out is a popular way going forward, but nothing is as simple as it first seems.
    The interesting part is figuring out what the optimal “unit of scale” might be. Processors are getting faster, storage is getting denser, etc. — what’s the ideal building block size for a particular use case?
    Make the building blocks too small, and you’ll have one set of ugly issues to go deal with. Make the building blocks too big, and there’ll be another set of crocodiles to wrestle with.
    As always, the devil is in the details …
    — Chuck

  2. Martin G says:

    True Chuck but arguably, IBM should be very good at this; they’ve been building scale-out compute architectures for a very long time. And if they hadn’t spent so much time, energy and money competing with themselves internally; they might have built a decent scale-out storage architecture some time ago?
    Don’t worry tho’, I still can’t see them actually manage to find the same hymn-book; let alone sing from it.
    And when are we going to see scale-out Celerrion?

  3. Chuck Hollis says:

    Well, depending on how you define things, the existing Celerra does a reasonable degree of scale-out for many practical use cases, although it’s not the uber-scale out model you’re talking about.
    The front end scales independently to 8 nodes (data movers), the back end arbitrarily scales with your choice of EMC storage (CX, DMX, V-Max, etc.).
    Other products in the portfolio that are customer visible today that do a better job of scale-out along the lines you are discussing would be Centera and Atmos.
    So, no direct answer for you!
    — Chuck

  4. Martin G says:

    there are times when I really admire your brass-neck…you actually made me laugh with the suggestion that the current Celerra could be considered as scale-out NAS! I take my hat off to you!!

  5. marc farley says:

    I thought Steve Legge spelled his name without an “e” at the end. I’m probably wrong. BTW, he is a great guy, one of my favorites in this industry. You really should invite him for a #storagebeers sometime.
    Thanks for giving 3PAR a mention. I want to make sure readers don’t misunderstand about our capabilities and aspirations. 3PAR’s architecture set the standard for this approach years ago and long before IBM and EMC introduced their versions. 3PAR has the largest number of “scale out” arrays in production and we aspire to keep it that way.
    I like Chuck’s point about right-sizing the architecture for scale out. Storage systems designers have to make a lot of tough decisions that impart both strengths and weaknesses to their architectures and products. Like so many things, it is much easier to do something poorly than to do it well.
    I think IBM’s shortcomings here have come more from lack of consistent leadership in their storage business. People have come and gone with some regularity. Steve Legge has been there for quite some time, but his organization went from being independent to being part of the systems business and with too many top level changes. It makes it very hard to stay on course when a new course is the order of the day. The company might be too big and too political to be effective.

  6. Chuck Hollis says:

    My apologies, Martin, I wasn’t trying to be a “brass-neck” (whatever the heck that might be).
    Again, Celerra isn’t exactly what you’ve presented here in terms of your description scale-out — I think I made that clear.
    But, you’ll have to admit, the ability to independently scale back-end and front-end *does* give it some small measure of scale-out-ness.
    Although I wouldn’t put it in the same architectural league as, say, the SONAS offering.
    Sorry if I overstepped some boundary …
    — Chuck

  7. Martin G says:

    Actually, you are right; Steve does spell his name Legg; well I’m fairly certain you are right. We keep meaning to have a beer out of the office but we’ve met a number of times. He is a great guy, actually most of the storage guys tend to be great guys; especially if they are not overtly selling you stuff!! There’s very few of you that I wouldn’t be willing to share some beer with.
    I think if IBM are ruthless and rationalise their product lines and go with a clear scale-out vision; they could do very well.
    And I’m sure there will be some discussions about whether 3Par is or isn’t scale-out storage. It has some aspects of scale-out storage for sure…

  8. Martin G says:

    It was an affectionate comment Chuck…at times, one just has to admire your cheek. There was nothing nasty meant by it!!
    BTW, surprised you didn’t mention Celerra MPFS…but then again, so very few people in EMC do or even know about it!!

  9. marc farley says:

    The 3PAR design is a cluster. If clusters are somehow excluded from your definition of scale out, then we aren’t. If scale out includes things like federated storage, then who is? (and who won’t be?) I think we will have problems with the definition. Its too convenient to wave a hand and make decrees.

  10. Martin G says:

    I’d expect genuine scale-out to scale over a petabyte; InServ currently doesn’t at present. And federation is probably the next big buzz-word in the storage world.

  11. Hector Servadac says:

    Martin, what do you think of solutions like this? Scales Out, not a storage device but a cluster with IBM’s GPFS:

  12. Hector Servadac says:

    Ok,you talk about NetApp & IBM and I think it’s a love-hate relation. NetApp was carrying distributors to IBM N-Series customers & prospects and that kind of misbehavior slows down N-Series sales. Then they smoke the pipe of piece, but still a bad taste in mouth of some salesmen.
    IBM is selling NetApp as a NAS appliance and nothing more, despite virtualization features, SAN capabilities, etc. If you compare in price/performance, DS5XXX is half the price than a N-Series with same IOPs in a SAN environment. Which one you’d chose?

  13. Martin G says:

    IBM needed a viable NAS solution; they didn’t have one five years ago; everything they’d tried was quite frankly rubbish. They knew that this was a growing market, so they looked to partner with some one; fortunately NetApp also needed a large credible partner to help them grow in stature, not capability. I also suspect IBM considered buying them at that point.
    Now IBM look that they are building a viable NAS solution; one suitable for the challenges of today based on their own proven technology. I think IBM are building an escape motorway; they are on the slip road at present but ready to vanish off into the distance.
    Scale-Computing have an interesting proposition but it’s really hard for a smaller vendor at present to get any kind of traction. The market is still very much in flux, smaller vendors are struggling but if they stay in their niche and aim at a specialist sector.

Leave a Reply

Your email address will not be published. Required fields are marked *