The Other Side Of Bricks

Lots of interest in Xiotech's ISE announcement last week.

Many of the industry pundits found their approach, well, interesting to say the least.  One industry curmudgeon (I loathe to call him an "analyst") was almost falling over himself to praise Xiotech for what they're doing.

Now, I don't know much about their product (other than what I've read), and -- of course -- nobody can actually buy one yet to see what it's all about, but I found myself explaining the pros and cons of such an approach.

Of course, it really doesn't matter what I think, does it?  Eventually, the marketplace decides these things ...

Innovation Is A Wonderful Thing

Just when you think storage is getting boring, someone comes along with a (purported) fresh approach to things that make people sit up and take notice.  I love this -- it leads to all sorts of interesting discussions.

And that's exactly what Xiotech did last week when they announced ISA.

Major claims:

• great IOPs, mostly through the use of a smaller, high-performance 2.5" drive from Seagate
• magic, self-healing disk juju (not a lot of talk about how this is done)
• brick-like subsystem that's supposed to be easier, cheaper, etc.

Let's look at these individually, through a slightly jaundiced (and aging) pair of eyes ...

Great IOPs

Smaller disks that spin faster?  I'll grant you that -- given their approach -- they probably can drive more IOPs that a similar number (or smaller number) of bigger or slower drives.

No free lunch, though -- a given capacity will cost more when offered on 73GB 2.5" 15K spindles than, say, something less exotic.  How much more?  We don't know yet.

From a tiering perspective, they'll be sandwiched between more prosaic 3.5" drives and slower versions of the 2.5" drives from below, including approaches involving short-stroking, e.g. just using the premium real estate on a larger or slower drive.

And, of course, from above, they'll have to compete with the new enterprise flash drives, which offer considerably more performance than any rotating media, but at a higher cost.

No obvious answer as to what's the "right" approach here.  If it's only a constrained amount of data that needs to go fast (think transactional databases), I bet enterprise flash drives get the nod.  If it's a mix of things, maybe a short-stroking approach and using the remainder for something accessed only occasionally might be the right approach.

And, let's not forget, those uber-speedy drives are available to everyone in the industry, including EMC.

Magic Disk-Healing Juju

Many of us were most skeptical of these claims for a couple of reasons. 

First, Seagate (and other drive vendors) expose all of their capabilities to large customers, like EMC.  You might argue that different vendors don't take advantage of all the techniques that are potentially available, but -- at least from EMC's perspective -- we've looked at just about everything, and implemented what made sense.

Second, I don't think it'd be in Seagate's best interests to do something "special" for one vendor and upset everyone else.  It just doesn't make business sense, does it?

Third, we've got hard data that says that the vast majority of "storage errors" have very little to do with the disks themselves, and have to do with more prosaic things like HBA firmware, dual pathing, et. al.

I'm not saying that Xiotech might have a clever approach to this that no one else has thought of, but -- based on everything I've read and heard -- this case has yet to be made. 

And many of us will be waiting breathlessly for more details ...

The Economical Brick?

Every so often, we hear tell of self-contained storage bricks that make scaling easy, cost-effective, etc.

But, if you think about it, the economics don't really work when you start adding up lots of bricks.

Each brick needs its own power supply, processor, firmware, environmental controls, etc.  Fine, no problem. 

But when you start using lots of bricks, there's a lot of redundancy -- e.g. a bigger processor could support more drive more cost-effectively, a bigger power supply could do the same, a larger enclosure could be more cost-effective, etc.

One of the most vibrant discussions in storage hardware design is how to build modular storage architectures that span a wide a range as possible for a given set of use cases.

And, sad to say, there's no "one answer", which one of the reasons why we have CLARiiONs in different flavors, DMXen in different flavors, and so on.

Disruptive?

No, not really, not so much. 

Innovative?  Sure, I'll give them that.  They've chosen a target market that they're hoping will emerge (e.g. very high performance storage), and built their architecture around that.

And, just to be clear, it doesn't matter what I have to say about all of this.

Customers eventually decide.

Comments

[Arne Rohlfs]

Interesting response to this blog post: http://www.drunkendata.com/?p=1679

[Chuck Hollis]

Hi Arne -- yeah, someone sent me the link, and I read it with a bit of amusement.

Jon's got an interesting take on things usually, doesn't he?

Personally, I see most of his value as entertaining in a certain sort of way.

I, for one, don't want to get into an extended debate in all the areas I think he's wrong, etc. Too big a job, and -- besides -- what's the payoff?

Thanks!

[Chuck Hollis]

Oh my ... Bill Bonnin sent along this interesting link, uncovering a whole angle to this thing that I hadn't expected ...

For you conspiracy theorists, good reading!

http://www.techconfidential.com/vc-ratings/money-in/seagate-gift-keeps-on-giving.php