When Dell Went Whisky Tasting

NetApp names internal software versions after truly terrible American beers, and unfortunately for those with more refined tastes, not the far superior British variety. (I am sure I will pay for this comment, but in my defence, anyone who has tasted both in a single sitting will know it to be true.)

Dell/EqualLogic name theirs after whiskys. Scotch to the uninitiated, and please note the spelling of whisky. No E. And how to pronounce these whisky names before we start. Some are straightforward, but, oh, the mangling some get....

It's a Macallan?

When Dell bought EQL, my surprise wasn't at the purchase itself. It made sense for Dell to get out from under EMC, and its own range of systems were uninspiring. What surprised me was the price; $1.4 billion. 

It's very large indeed when you look at what an EqualLogic PS Array actually does. At first glance it looks like a smart SAN technology -- definitely smarter than the EMC CLARiiONs they sell -- and perhaps that's why Dell got over-excited. 

An EQL box virtualises storage; the LUNs aren't "real" LUNs, to borrow EMC's daft mantra. Then, on top of that, snapshots, clones, async replication and thin provisioning, with a touch of scalability thrown in the mix. Over iSCSI.

I think that's the entire feature list... Unfortunately, it's nowhere near as smart as a NetApp system. Point in case; deduplication. 

Or perhaps a Laphroaig...

There's a buzz about deduplication right now, driven by some very pressing business needs that require companies to retain existing data, and contain the explosive growth of new data that they're generating. (Or having generated for them; the average intertubes user is generating megabytes a day, and there are mbillions of them.)  .

So it seems strange that one of the missing Dell/EQL features is deduplication. With storage virtualisation, you'd think that it would be there -- or at least on the roadmap.

Maybe it's a Dufftown.

Dell/EQL is stuck between the devil and the deep blue sea.

EqualLogic engineers aren't going to introduce deduplication any time soon. There are some very specific technical reasons related to the smallest level of granularity that a PS Array can virtualise.

The unit of management is a tens of megabyte page or chunk, and block deduplicating them (this is a just a block-based SAN, remember) is only going to give results if there are duplicate tens of megabyte chunks. Oh, and across multiple arrays too. I reckon...

It's a Knockando. 

The next Dell/EQL release name? If it's deduplication you're after, Knockando. 

Sliante!

Comments

[Val Bercovici]

Why am I so thirsty after reading this? :)

[Alex McDonald]

Don't slake your thirst with a Glen Kinchie, whatever you do. Just think back to the last time you and I tried it :-)

Alex, looks like you've been inhaling sterno again. I'm sure you meant to say that EqualLogic will have de-dupe about the same time Netapp has a competitive iSCSI product.

[Alex McDonald]

Interesting comment, since NetApp already has iSCSI -- at a competitive 10GbE too.

The problem with the iSCSI figures for market share (which is what I presume you're referring to) is that they don't show that many iSCSI-only systems are bought as back-ends for NAS services. Nearly *three quarters* of EqualLogic systems would appear to be sold for that purpose.

Which begs the question; when is Dell/EqualLogic going to do NAS, if that's the market? Not any time soon, I reckon. It would badly dent Dell's server market.

I may turn off anonymous postings, btw. I really don't see why people want to remain anonymous on blogs like this. I'm not that desperate that I would want to track you down and breath Sterno fumes on you.

[Marc Farley]

Alex, the previous comment was from me, Marc Farley from Dell EqualLogic. The comment above wasn't intended to be anonymous as I had logged in with TypeKey and rec'd the message "You are currently signed in as Marc Farley". You'll have to figure out why the system made me anonymous, it's your blog. I see now that you've sort of fixed things - because after submitting this post, I rec'd an error and am being asked for my credentials, even though my typekey profile provides that stuff on other blogs like Chuck Hollis' and Nicholas Carr's.

I'm pretty sure we aren't selling 75% of our systems as file server front ends - unless you count lots of guest systems running on VMware in that total. Then maybe.

NAS? Really? Why would we need that if we are selling so many systems as "back ends" for file servers? Maybe its Netapp that needs unencumbered block storage instead.

[Alex McDonald]

"The unencumbered Dell/EqualLogic PS Array"

I like it! "Can't deduplicate or do NAS or [long list of things it doesn't do]" I can now explain with the one word.

Thanks for the bug report; I'll pass on to the web admin team. I'm having problems posting comments too.

[Val Bercovici]

Hi Marc, and welcome to NetApp's little blog corner of the storage world.

Please indulge us and elaborate on what exactly "encumbers" our storage arrays today?

[MarcFarley]

Marc here again. By unencumbered I mean the way that block storage is processed through the file system layer of a Netapp filer. I'm sure it was done to ensure consistency with filer snapshots, but it doesn't help performance much.

[Anon]

Netapp are to busy looking at others products and charging silly money for their toys whilst Dell/EMC/Equallogic continue to sell bucket loads of storage, just look in any data centre, I dont see NetApp arrays, actually the last one I saw was...err I have never seen one.

Ibm yes, HP yes, EMC yes, NetApp no..so Alex sorry mate but you can continue writing in your blog as I guess NetApp arent very busy doing much else :)

[Val Bercovici]

Thanks for playing along Marc, and for taking the bait.

The urban legend that NetApp's block storage interface (for FC-SAN's or iSCSI, or soon FCoE) somehow sits on top of a filesystem is merely anecdotal, with no bearing in fact or reality.

Block I/O's are at a peer layer to NAS filesystem semantics on a NetApp FAS/StoreVault array. That means they cut-through right to the virtualization (aka array volume mgr) later with integrated RAID. Snapshot integration happens at that lower layer, and likewise has nothing to do with NAS filesystem semantics.

Ironically, the independent benchmark evidence such as SPC-1 proves how our disk virtualization layer actually accelerates performance rather than encumbering us with any "overhead". And remember, thin provisioning is a native attribute in this implementation and is included in all our performance results - as are snapshots. Sorry to say (for our competitors that is :) we'll be adding dedupe to our performane results soon too!

In fact, you need look no further than the NetworkWorld iSCSI review we won today where our low-end solution handily outperformed your high-end PS series and all other iSCSI comers (see slideshow on their website) for proof NetApp customers can enjoy all our robust features without encumbrances of any kind :)

http://www.networkworld.com/slideshows/2008/072808-test-iscsi.html (See Slide 8 for the performance results)

[Alex McDonald]

Anon;

I've never been to China, so it doesn't exist.

[Hugh]

I love how NetApp big up their own de-dupe.
Then go and buy Data Domain, as the NetApp de-dupe sucked, and until recently played havoc when a snapshot happened during de-duplication time.

And who needs de-dupe. Our EQL boxes already have 200% the capacity over a NetApp box, as it doesn't have WAFL stealing 50% of the storage straight out of the box, for 50% of the price. So EQL is about 25% of the price for usable disk ;) LOL

I have all 3 appliances (Data Domain, NetApp and Equallogic), but the NetApp FUD is just as bad as any from EQL.

[Alex McDonald]

@Hugh

Thanks for your comments.

Using best practices for dedupe might help you with understanding how snapshots and deduplication interact; see section 4.1 on page 20. There is no measurable performance effect on taking a snapshot when deduplicating, as a snapshot does very little work indeed.

I don't know where you get the 50% WAFL overhead from. Are you sure you're using the right commands to get the usable space on your systems?

Nor do I understand how you can get 200% of the capacity over an equivalent NetApp system. Or are you saying you have twice the raw capacity?

A call to support may help if you are having problems in managing your NetApp system.

[Hugh]

Hi Alex,
I do understand how the snapshots and de-dupe work. It wasn't performance, but crazy amounts of space used that was the issue. I think 7.3? of OnTap fixes this. I know its due to be fixed, but we also got better ratios with Data Domain. Well done for acquiring them, as their IP is clearly very good.

I was being tongue in cheek on the WAFL overhead. But customers know that if you buy 20TB of NetApp that is "raw", and you probably get 50-60% usable for actual work.
Whereas on EQL you get the 100%, as there is no WAFL.
Humorously, even the screenshot here:
http://blogs.netapp.com/exposed/WindowsLiveWriter/NetApp-Synergy-120-drives%28disk-view%29.png

On your own NetApp blog shows the lies.
"Marketing RAW" LOL But what you can actually "use" is 13TB not the 20TB you would be sold. We've been caught by this before ourselves. That was before we became clued up on storage.

So yeah, 200% means we get effectively double the space for half the price, hence my figures :)

Anyways, good luck with your products.

All the best
Hugh

[Alex McDonald]

Hugh

Your dedupe ratio will vary by data type, and as it's post-process, under very high change activity the amount of deduplciation may fall considerably. The best practices covers that.

You actually get 66%; and the screen shot you reference clearly shows that. Every manufacturer states drive sizes in "marketing" sizes (base 10), including Dell/EQL, and every manufacturer then

1. "rightsizes" the disk
2. takes out overhead for metadata (which EQL has too)
3. present the rest as usable space.

A 1TB (base10) EQL disk will be considerably smaller once this is done. And without dedupe, that's all you can use too.

I object to the word "lies". We always size based on the user's usable requirement. Otherwise, why would we have such a sizing tool that we use with customers when assesing their storage needs?

[Alex McDonald]

A follow up;

Given a 16TB (marketing) PS array (16 disks), set out as RAID50 (striped RAID5), then the sums per system are:

2 parity, 2 spare, 12 data = 75% or 12TB (marketing)

Then rightsize them (non-marketing numbers) and you'll be pretty close to 70%, compared with NetApp's two thirds. How that equates to your "double the space" comment is beyond me.

[Hugh]

Hi Alex,

Whilst I get my Equallogic numbers together, perhaps you can help me show you the real NetApp numbers.

I downloaded Synergy, but I can't seem to get it activated. The software isn't proxy aware, so I plugged it into a plain internet connection and it still wouldnt work.

So, in the latest aggregate I purchased, we have a shelf of 14 x 828G disks in RAID-DP.

This was sold to us, the end user as a shelf of 14 x 1TB drives. So, 14TB.

Now you tell me how much I can "actually" use of that number. I will do a like for like comparison with Equallogic if you have the determination and honesty to induldge my fancy :)

[Alex McDonald]

Hugh, let's do this offline, and we can do all the detail stuff before we post the results back here. I'll send you an email to the contact details.