Storage Shenanigans

I'm taking a break from morbidly watching the global economy reverberate, as I saw an interesting item pop up in email a few hours ago.

I should really leave this one alone, but I can't.  I find it just too amusing at a certain level.

Where To Start?

I think everyone who reads this blog regularly knows that I have ongoing issues with NetApp's general conduct in the marketplace.  It's more than just a a competitive thing,  And I think most readers followed the recent usable capacity debate from a while back.

Now, both themes have come together in an entirely new offer from the good people at NetApp with a grabbing headline: "50% Less Storage For Virtual Environments Guarantee".

Since I'm perhaps too close to all of this, I'd invite your esteemed opinion as to whether this is a truly useful offer, or yet another gimmicky marketing stunt from NetApp.

The Pitch

The structure of the deal is that you'll presumably use 50% less storage in certain virtualized environments, and -- if you don't -- NetApp will give you extra storage to "make good".

If I wasn't such a storage geek, I'd be tempted to think "wow, what a great deal!".

But then I went looking through the details, caveats and exclusions, and came away thinking that this was the kind of promotion targeted at IT buyers who might not have the time to read the fine print. 

Personally, I detest marketing initiatives that prey on people's lack of familiarity with certain topics.

So, what's my beef?  Go look at the fine print.

-- You must use RAID 6 as compared to  "traditional" RAID 10.  Now, we can argue about the pros and cons of each, but most storage people would not consider them directly equivalent in terms of protection or performance.  Never mind that you can get RAID 6 on all sorts of storage arrays today.

-- No databases, email servers or other transactional applications need apply.  Wow.  That's a whole lot of VMware use cases that are not part of this "guarantee".

-- No more than 10% of data that can't be easily compressed -- jpgs, powerpoints, XML data, etc.  There goes a lot of the files people actually use.

-- You have to be willing to dedupe primary storage.  That's a can of worms in itself, isn't it?  Not to mention that it's a post-processing scheme, so you'll need extra storage lying around for that as well, which presumably doesn't count as part of the offer.

-- No snap reserves, or -- at least -- no counting any space you set aside for snap reserves.

-- must use thin provisioning without LUN reservations.  Never mind that this is a pretty common feature on many storage arrays today.  And there's no counting any space you might have around to protect yourself against running out of space.

-- excludes performance-sensitive environments that might require a certain number of spindles.

-- You need 10 or more virtual machines doing pretty much the same thing, all in the same flexvol.  Not clear how one measures this, or if you exclude everything above, whether this is even close to a realistic situation.

-- full suite of NetApp pre and post services required

-- detailed claim process if you think you haven't achieved your 50% space savings.

-- the proposed remediation is extra drives to "make good".  No mention on extra shelf space, power, cooling, etc.

I could go on, but by now you might guess why I'm sharing this with you. 

I feel like I'm watching one of those local TV ads watching Crazy Eddy peddling used cars on Sunday morning.

If You Don't Mind, A Couple Of Questions

• First, do you think this is a serious offer that has economic merit?
• Second, do you think this is a serious offer with technical merit?
• Third, do you think that a significant number of customers could benefit from the "guarantee" as presented?

I'd be interested in any and all opinions -- that is, from people other than NetApp employees.

[update: some of the bloggers at NetApp took this last comment *very* personally.  I guess I mis-spoke.  I'm sorry.  Of course, I'd like to hear from them, as would we all.  I had no doubt they'd react vigorously as they always do, I just wanted to hear from other people on this one.  And I have, so thank you!]

What I Think

I do have my opinions, don't I?

I think that NetApp is under the gun for historical poor space utilization.  We can argue about what NetApp is doing today, but many people will attest that they've got a lot of filers out there with very poor space utilization.

Past sins are coming back to haunt them.

I also think that NetApp is losing the battle for VMware storage "mindshare" against EMC and others.  If you were at VMworld, you probably saw this to a certain extent.  If you subscribe to all the market share feeds like I do, you'd also be seeing the effect.

Put the two together, and you have the "perfect" marketing promotion, don't you?

Just like a political campaign, the candidates try to highlight the differences in their approaches.  Let me highlight a key philisophical difference here: if someone at EMC were to propose this kind of marketing stunt, I'd do everything in my power to shut it down as a Really Bad Idea.

Why?  Because it doesn't do most customers any real good, if you think about it.  Sure, it's an attention grabber -- I'll give them that -- but, to me, it says a lot about how they view their prospective customers. 

And that view is not too flattering, IMHO.

Update (later in the day)

I had a few moments today, so I went over to Nick Triantos' blog at NetApp, who was touting the new program as expected.  And it was an interesting (though civil) discussion, to be sure.

Nick's first claim was that RAID-DP was equivalent to RAID 10 in terms of performance, and offered up his homegrown SPC-1 results as proof.

I countered that wasn't a reasonable position, as follows: take any decent workload on 8 spindles.  Now add two to get RAID 6, for a total of 10 spindles.  Take that same workload and run it on 16 spindles as you would with RAID 10.

I find it very hard to believe that the performance would be equivalent, don't you?  A similar argument could be constructed around availability, but that's not even fun to do.

Nick's second point is that "NetApp stands behind its guarantee" and "we're putting our money where our mouth is" and similar aggressive-sounding statements.

I offered that, given the long list of caveats, exceptions and other wormholes, there isn't a whole lot left to stand behind.  I also mentioned that I was just waiting for the press release in a few months that NetApp had never had to pay up on its "guarantee"!

I pointed out that a lot of standard arrays support things like RAID 6 and virtual (thin) provisioning.  What's the big deal here?

So it must all be about the dedupe of primary storage.  Unless of course we're talking about database, emails, powerpoints, images, XML files, busy data and much more that's "excluded".  And, of course, any environment desiring some level of performance.

Funny thing, though.  Take any 50 application environments.  Virtualize them.  The data is pretty much the same, right?  So why is this only available on virtualized environments?  Seems to me that it would work pretty much the same (except for a long list of caveats) on physical and virtual environments.

Nick, in a moment of probable testosterone poisoning, challenged EMC to a similar "guarantee".

Despite the fact that EMC abhors this sort of gimmick, I offered that for many years EMC has offered a storage optimization service where we come in and look at process and procedure with the goal of reclaiming underutilized storage.  Sometimes we do this on the basis that we'll find more storage savings than the service costs, so you don't pay unless we find the pony.

The ground rules for this is that we can't harm service levels, or availability levels, or force IT into dramatically new processes.  I should point out that EMC does this on anyone's equipment (EMC and others) and does this on equipment of any vintage, not just brand new boxes sold after September 30th, 2008.

I also noticed that the broader NetApp Blogging and Cheerleading Corps (NBCC) have strangely not picked up on this, reinforcing my belief that this sort of "marketing promotion" (I'm being kind here) was not endorsed by the broader technical community.

[end update]

And, While I'm At It

Larry Ellison's announcement at Oracle getting into the storage biz resulted in many industry tongues wagging as to whether all of us storage guys should be watching our back.  Trust me, if you've been in this business as long as we have, you're always watching your back.

Someone sent me some interesting math based on the published pricing for this particular specialized storage device.  Please take this with a large grain of salt, since we're talking list prices, and there's not a ton of detail as to how the configuration splits out.

That being said, it looks to be about $4/GB raw, $8/GB usable.  Wow.

But the real gem is the software pricing, which looks like $10/GB raw, $20/GB usable.  Double wow.

Draw your own conclusions.  My take is that (a) they're not entirely serious at these prices, at least from a storage perspective (b) this is a purely defensive move against folks like Teradata and Netezza, albeit an ineffective one, and (c) you can see where Oracle wants to put the value: in software.

I give this a few weeks for the post-OOW marketing spin to play out, and it'll most likely sink below the waterline to join past Oracle initiatives in the deep blue sea ...

Courteous comments always welcome!

Comments

[puff65537]

-- You need 10 or more virtual machines doing pretty much the same thing, all in the same flexvol

This is actually a pretty common scenario due to pathological corporate politics, tight IT budgets, and every department with a half way competent techy wanting to control their own servers. IT/IS simply says F*$% it and gives everybody their own VMware image to run themselves. De-duping just the OS files should be a huge win.

[Chuck Hollis]

Yeah, fair enough, but do you see this resulting in 50% less storage utilization?

Now, I see your point -- imagine a whole bunch of little "crapplications", each with a tiny bit of user data, and all of them requiring the exact same binaries to run.

But does this mean -- in the real world --you need 1 array instead of 2? That's where I have the problem ...

[Martin G]

Okay, they've bounded this in a such a way, it should be hard to fail. Hardly a stretch target to be honest. But I do think it is an interesting way to highlight the use-cases for primary storage dedupe; I can see it having some great potential in development/test environments where I have many copies of the operating system, Oracle binaries, Webserver, Application Server etc.

I probably wouldn't want to rely on A-SIS in my production environments and it probably wouldn't get to run anyway as I have things like batch-runs which often mean my arrays are busier overnight than the online day, so I don't have spare CPU cycles to play with.

And with vStorage Linked Clones coming along the track; NetApp do need to highlight a potential competitive advantage whilst they still have it.

Its probably all goodness as it gets some of the ideas into the open and lets them be debated now. I hope that in spite of what is a fairly inflammatory marketing campaign that we might get some interesting debate.

But I am fairly certain that in a few years time; we will see primary storage dedupe becoming a fairly standard feature (just like thin provisioning) but only for some use cases.

[Chuck Hollis]

Agreed on all points, thanks Martin!

[puff65537]

> Yeah, fair enough, but do you see this resulting in 50% less storage utilization?

Depending on how draconian central IS/IT is in making sure patches/service packs are installed, I could see 80% savings (assume a standard of XP, 2K3, and RHEL4) and that would take you from 2 to 1 array and give users more room for their own insular little apps. I would bet one previous place I worked would be all over this, with everybody on board. Bean counters furiously counting, all the little empire builders willing to stick to the standard in return for more buildout room and IS, still short staffed even with all the help, wondering if they can fob off basic admin to all departments.

[Chuck Hollis]

LOL!!!

Get a take on Steven Schwartz' blog:

http://thesantechnologist.com/?p=122

[Dave Mcdonald]

On the contrary, I see this as a bold move on NetApp's part. No truly binding offer comes without fine print. This is exactly the kind of thing that gets a CFO's attention.

Besides, as a customer what have I got to lose? Either I save a lot of capital and operational expenditures deploying less storage - or I save just capital dollars via free upgrades from NetApp to satisfy my requirements.

Sounds like a win-win to me. I for one hope EMC and other storage players jump on this bandwagon to prove their claims in the most tangible way - by putting your chips on the table where everyone can see them!

[Chuck Hollis]

Dave, I stand corrected! I guess there IS an audience for this kind of marketing stunt!

BTW, is this the same Dave McDonald who works at Peak, the NetApp reseller?

-- Chuck

[Mike Shea]

This is a first. The first time I've seen someone actually attempt to characterize a guarantee as a 'bad idea' - unless you are talking sports....

Amazing.

(Yes - I work for NetApp)

[Chuck Hollis]

Mike, most guarantees imply something of value.

This one doesn't, does it?

It comes across as a marketing gimmick, not an honest attempt to put the customer first.

And I'm not the only one with this opinion.

[Paul P]

Let's compare with some EMC quotes.

http://www.emc.com/collateral/hardware/data-sheet/h5527-emc-clariion-cx4-ds.pdf

"Low-power SATA drives use 96 percent less energy per TB versus 15k FC"

Wow, SATA drives using only 4 percent of of the 15k FC variety. Is this a normal expecation in normal evironments? Is the fact that EMC do not offer any fine print on this statement and no offer of a money back gaurantee make it less gimmicky or 'more' true?

Let's try another one:

"The CX4 boasts up to twice the performance of the previous generation"

Are you stating that CX3's are twice as slow, again, I didn't see the fine print so maybe I am missing something here.

[Steven Schwartz - The SAN Technologist]

Thanks for the linking Chuck! When I read the NetApp release, and did the basic math, I just couldn't believe it. Forget about FlexClones, Deduplication, Snapshots, etc. If you take best practices for a RAID10 system vs. NetApp's recommendation for a RAID-DP system, and compare usable capacity, there is a 50% "savings" just in the different RAID choice.

I used 4x 14 drive trays (500GB drives) as my example since NetApp required a minimum of 10TB usable. If you configure 56 drives in RAID-6, closest available RAID configuration in comparison to RAID-DP, and assumed no spares since each raid set is dual failure protected, you end up with 26TB usable. If you take the same number of spindles, assume standard best practices of a single HS per tray for RAID10 configurations, you end up with only 13TB usable. So working math backwards, you only need 28 spindles with RAID6/DP to achieve roughly 12TB-13TB of usable capacity.

So all that is being committed to, or rather the "guarantee", is that with RAID-DP you need less disks then with RAID10. Now let's be honest, in MOST IO profiles, RAID10 and RAID6/60/DP, DO NOT have the same performance profile, so the original baseline, while on capacity might be a savings, number of spindles and performance won't be!!!!

In either case, I'm glad I could bring some laughter to what I considered a pretty comical marketing ploy.

[KPC]

I think NetApp don't need to do this kind of "guarantee" kind of marketing. The product is any way of high quality for the enterprise. The NetApp services and support has really improved. So capitalize!

EMC has the best storage marketing today and
NetApp marketing team should take a leaf out of EMC and learn what to pitch.

If you can save half the capacity on NetApp storage for VMWare, SO WHAT?

NetApp should take some other route like EMC ILM/Cloud Computing or HP-Database Machine etc...messages.

[Chuck Hollis]

Hi Paul

Obviously, you've got a different perspective on this than many people.

But let me see if I can help.

There's a difference between comparing two technologies, and "guaranteeing" that all customers can see this in their environments, and then lard in a whole pile of disclaimers which render the "guarantee" valueless, which we would never do.

What we do is say that there are specific use cases and tradeoffs, and take the discussions from there. You may not appreciate the difference, but it's pretty significant.

Ditto for performance claims -- no "guarantee", as we all know your results may vary, and then we get very specific as to what and where. Again, no marketing stunt.

Thanks for writing!

[Steve Marsden]

Steven

That is priceless! Thanks for doing the math. I guess NetApp could have made their offer a little less complicated if they had done the maths themselves.

But hats off to their Marketing team right? It is fairly obvious this offer has been prompted by Chuck's blog a few weeks ago. They definitely want people to believe their hype. The sad thing is that some people will buy into it. Great - RAID6 (or DP) uses less spindles - and we'll give you money if this isn't the case.

Their other marketing coup was of course A-SIS itself. Completely impractical in terms of its scheduling and the one area where you might choose to use it - file system data - usually runs up against the maximum flaexvol size limits - but a great marketing message.

All very amusing.

[Dave Mcdonald]

Hi Chuck - Dave again.

No I'm not the D.M. that works for Peak, but I do represent a large systems integrator which sells and uses HDS, EMC & NetApp.

While I'm not a big fan of their hardware (3100 shows promise) I have to side with Curtis Preston that NetApp's software is second to none - especially with regards to offering more usable capacity than all the others by a country mile.

I have seen all 4 major components of this guarantee differentiate themselves vs your gear and HDS' in real-world PoC's and production.

RAID-DP is easily as fast as RAID 10 without the mirroring overhead, FlexVols implement thin provisioning with no extra I/O later, snapshots don't mess with response times and the new A-SIS icing on the cake accomplishes eye-catching dedupe ratios for VMware.

DMX's and Tagma's have their place in this world, especially for Mainframe or Superdome-class requirements, but in the x86 environment NetApp has a strong value prop.

[Chuck Hollis]

Wow -- you're "on message" for NetApp!

[David G.]

Hi Chuck,

Just a quicky - NetApp don't use RAID6, and while you may be trying to "keep it simple", the only similarity between RAID6 and RAID-DP is the parity overhead by space - the data protection and performance characteristics are very different.

On that topic though, RAID 1+0 and 0+1 both have the possibility of data loss after a failure of only two disks (for RAID 1+0, two disks on the same pair, or for RAID 0+1, a disk on each stripe set). RAID 6 and RAID-DP need to have three disks fail under any circumstances to lose data...