EMC Atmos (Maui) Is Here

Today ends most of the speculation around the often-speculated-but-seldom-seen "Maui" product, now offcially known as Atmos.

Before we get to the usual skepticism that accompanies any unique or novel technology, let's take a look at what EMC has brought to market, what problems it tries to solve, and some of the underlying technology concepts.

Ready? Let's go!

A New Category -- COS -- Cloud Optimized Storage

As part of this product introduction, we're going to be introducing a new term -- COS, or Cloud Optimized Storage -- to the industry lexicon.

Why? The traditional storage taxonomy doesn't do a good job of describing what Atmos (and, presumably, future solutions from other vendors) actually does. As you'll see shortly, it isn't SAN, NAS or even CAS.

So, what makes "cloud optimized storage" so different? The use of policy to drive geographical data placement.

OK, that's an extremely abstract concept, so let's build it up in pieces.

A Simple Model

Imagine you had a single data center with all sorts of juicy content that people wanted around the globe.

If you happened to be close to the data center, access times would be pretty good, right? But, if you're halfway around the world, data access could be slow, almost to the point where it might be unusable.

If it's a very popular piece of content (say, the latest political spoof from Saturday Night Live), just having one copy in your data center sitting on one spindle wouldn't be enough to keep up with demand, would it?

To deliver a reasonable user experience, you'd want to have (temporarily) multiple copies, ideally dispersed across multiple continents.

And then there's networking costs. Spending on big pipes to deliver essentially the same content from one side of the world to the other over and over again strikes most people as wasteful, not to mention expensive.

Global mobile phone operators have discovered that when there's a popular set of football (soccer) matches in Europe, people around the world want to watch them on their mobile phones.

That's a lot of redundant bytes being sent over very long wires. How about making a temporary copy closer to where people are accessing it?

So, as a starting point, let's imagine a global object repository, formed by multiple storage nodes scattered around the internet, all seen as a logical whole. Not really a file system in the traditional sense, although it could be presented as one if needed.

Applications load content from anywhere. They're now part of the 'global object pool'. When you load the content, you specify a policy, such as "gold" or "free" or "secure" or "pay per view" or "we think this is gonna be really popular" or "keep a certain minimum number of copies around for redundancy purposes".

That policy specification is dynamically interpreted by the Atmos environment. If something gets very popular, and access times elongate, Atmos can make multiple copies based on where the interest is coming from. And when the demand storm subsides, go back to a more cost-effective scheme.

The same mechanism can be used to implement, for example, authentication and digital rights management schemes. Or any other external logic that's triggered by either information access, ingestion, or external events (such as the passage of time).

As the definition of "policy" is rather open-ended and extensible, this approach leaves the door open to all sorts of clever applications in all sorts of surprising areas. Examples might include the capture and distribution of geophysical data. Or video-based training and certification. Or large-ish source code libraries. Or global software distribution. Or all sorts of video-on-demand models.

Use your imagination.

If the information objects in question are big (or there are a lot of them), and they're all being used around the globe, and it's going to be hard to predict when it's going to be popular, it's worth considering an Atmos-style approach.

What It's Not

As with any new technology, people tend to say "well, that's just like XYZ". Making those sorts of statements will be very hard to do when considering Atmos. In fact, most of us believe that this sort of approach is pretty unique in the industry.

So, let's go through the list of what Atmos isn't.

First, it isn't "clustered NAS storage", unless your definition of "cluster" includes geographically dispersed nodes, and your definition of NAS includes global object stores, rather than file systems.

Second, it's not a straighforward content delivery network as provided by Akamai and others. Not only is the functionality much richer with this sort of approach (driven by an extensible policy engine), but it can easily handle not only objects of considerable size, not to mention potentially several billions of objects.

Third, it's not something like a Centera. Sure, some of the object repository and metadata thinking can be found there, but this is an entirely different technology set and use case. With Centera, it's more focused on the preservation and retention of information, rather than global scale distribution.

Fourth, it's not really a new storage array, at least from a hardware perspective. The Atmos hardware is basically very dense, very cost-effective storage, with industry standard servers running the Atmos software. It competes well with others of its class, and has some interesting packaging innovations, but if you're a storage array geek like me, not a lot to get overly excited about.

And, finally, it's not the over-used "Web 2.0" storage as described by IBM and others. Those are still fairly traditional arrays, sitting in a single data center location.

Lots Of Coverage Today (updated periodically)

I'll be updating this section with all the coverage as it emerges today.

For starters, there's StorageZilla's excellent post, as well as Steve Todd's, not to mention coverage from StorageBod, Chris and a few others.  Chris Mellor at The Register checked in with an interesting take, as usual.  Even more from Network World and Tarry Singh -- both interesting!

And, if you're really curious, you can go back through some of my previous posts, and see all the bread crumbs I tried to drop along the way :-)

A few questions are coming in that deserve answers.

Was this an acquisition, or did EMC develop Atmos themselves?  Not that it really matters, but this one was a 100% organic development, although you can see some obvious DNA from earlier EMC offerings and initiatives.

Do you have any customers?  Yes, we have several.  As Chris Mellor points out, Atmos has been in customers' hands since June of this year.  They're names you'd recognize, but -- as usual -- they don't really want to be part of EMC's PR initiative :-)

Why did you wait so long to announce this?  Lots of reasons, really.  One is that we really didn't need to -- we had the product, we had customers who wanted to give it a try, so we didn't need to drive demand.  Another was a natural conservativeness around completely new technology -- we wanted to make sure the product worked as advertised, that we had happy customers, etc.

Is Atmos hardware-agnostic?  Yes, that's the design.  It runs well as a VMware guest, for example.  That being said, our experience with customers so far indicates a strong desire for hardware that's built for purpose -- especially at this sort of scale.  So that's why you see the nice arrays as part of the announcement.

Who are you competing with?  That's a tough question.  Other that a few home-grown solutions out there, we're not aware of any other vendor who offers a product like this for sale.  I'm sure that won't last very long, though. 

Typically, when EMC announces something like this, we go through a period of competing vendors doing the competitive trifecta (you don't need it, it doesn't work, and we'll have something better soon anyway), and then we start seeing similar offerings from other vendors, usually in 12-24 months. 

So, I'll be better able to answer your question in a year or so.

What about VMware's cloud strategy?  And does this really change anything for 98% of the people that are wrestling with traditional IT challenges?  Steve Foskett takes a skeptical view of all of this (probably the first of many to do so), asking fair questions.

He's right -- Atmos doesn't solve problems in today's traditional data center.  I don't think anyone at EMC ever positioned it that way, though.  Atmos also doesn't make a decent cappucino, in case that one comes up as well.

And, if you think for a moment about vClouds slinging guested virtual machines hither and yon, one does has to ask the question about how the information might follow it?

[Update on Nov 11 AM] 

Well, EMC's competitors have taken notice of what we're doing, and have predictably weighed in. 

CalvinZ over at HP thinks this is all about hardware, and has taken the time-tested approach of name-calling (proprietary, monolothic, etc.).  HP thinks Atmos is a traditional clustered NAS device, which is incorrect on several counts.  I tried to leave a comment on his post, but I guess I'm not welcome there any more.

Marc Farley (formerly of EQL, then Dell, and now 3PAR) weighed in with an unusually rancorous post that was more about me than the company or product.  I left a comment suggesting he might lighten up a bit; we'll see if it goes through.

It's not that I was expecting a congratulatory note or anything from these competitive bloggers, but -- really -- I was hoping they could do a bit better in responding to the Atmos announcement. 

It's pretty hard to intelligently respond to all that whining.

What Does All Of This Mean?

For those of you consuming traditional storage with traditional IT use cases, none of this will matter much to you in the short term. Sure, it's all interesting, but not many typical IT organizations are wrestling with the problem of global content distribution and logistics.

But, for a few of you -- and you know who you are -- this sort of approach will be inherently intriguing. You're running a global business. You're ingesting and distributing content from everywhere to everywhere.

All you can see in the future is more, more, more and still more.

And I think you'll be very intrigued by EMC's fresh thinking on the topic.

Comments

[Reuven Cohen, CTO Enomaly Inc]

A while back I came up with a term I described as the " Content Delivery Cloud" I think the approach of EMC's cloud optimized storage fits into this concept very nice.

Here is an overview of CDC
A Content Delivery Cloud is a system of computers networked together across the internet that are orchestrated transparently to deliver content to end users, most often for the purposes of improving performance, scalability and cost efficiency. Extending the model of a traditional Content Delivery Network, a Content Delivery Cloud may utilize the resources of end-user computers ("the cloud") to assist in the delivery of content.

More details here > http://www.contentdeliverycloud.com

[Stephen Foskett]

Thanks for that quick answer to my central question, Chuck!

It seems that any time EMC releases anything, observers want to position it as earth-shattering. But that's just not the case, at least in terms of customer scope. EMC has a wide range of products, from LifeLine to CX to DMX to Centera to Atmos, and while the whole pile covers the market, none is applicable across the board. It looks like Atmos will have a few very large customers, for now at least, and it looks like that's fine by you.

It does seem a bit out of place to call it "a new era for IT" on the product page, though. Cloud computing is a new era, but Atmos isn't, at least for IT in general.

Stephen

[Chuck Hollis]

Hi Steve -- agreed -- "a new era for IT" is a bit over the top! It's cool technology and all, but it doesn't solve world hunger, does it?

Thanks for commenting!

[Richard Brannigan]

Where is the software to let me run this in my VMware guest? The only thing I could download on your site was a PDF with a lot of expensive looking racks. I would suggest the following edit to whomever ghost-wrote this post, "it works fine as a VMware guest, but we won't let you do that."

[Joseph Martins]

While this is certainly different, the individual elements aren't new...at least none appear to be new to me and I've been in the information management game for more than a decade.

The news is in the integration. Intelligent geographical content distribution and caching combined with elements of content management, personalization, security etc...we saw some of the influence in the earlier days of ECM, but I'm certain no single company ever delivered on it.

We'll certainly continue to watch this closely.

[Chuck Hollis]

Hi Richard

Sorry if I disappointed you. I said "runs well in a virtual machine". I didn't say "and available for download to anyone who wants to".

Umm, there are no ghostwriters here.

Unfortunately, I have to write each and every word that appears here -- warts and all.

[Chuck Hollis]

Hi Joseph -- perhaps you're right, perhaps we've seen elements of all of this before.

I suppose we could say the same of most new or popular technologies in the marketplace today.

And, if you ever share this thought in front of an IBM employee, you'll learn that they invented it all ... :-)

-- Chuck

[Calvin Zito]

Hey Chuck,

As you probably guessed, we had some thoughts from what we knew before hand and what was announced yesterday. Those are in two posts: http://www.communities.hp.com/online/blogs/datastorage/archive/2008/11/10/pie-in-the-sky-with-atmos.aspx and http://www.communities.hp.com/online/blogs/datastorage/archive/2008/11/11/further-quot-day-1-quot-thoughts-on-atmos-maui.aspx.

The thing I'd most like to hear you address is the proprietary appoach you are taking with Maui that will effectively mean "vendor lock-in". The second blog entry is the one that talks about that.

Also noticably absent from your efforts yesterday is a single customer that has been using the "GA release shipping since June". I can't believe that if you've been shipping GA code since June that you don't have a customer who can talk about Maui. I could point to a couple of customers who've told us that Maui isn't ready for enterprise customers but I'm guessing they've signed some legal documents that prevents them from saying anything in public.

Congrats on a nice marketing job yesterday,

Calvin

[Chuck Hollis]

Calvin

I can't believe you're trying to compare HP's traditional clustered NAS offering with something like Atmos. Goes to my earlier theme about servers guys and storage ...

Different use cases. Different technologies entirely. Now, if you'd like to compare and contrast with something like, say, EMC's Celerra, we could have a decent dialogue.

Appreciate the "closed, proprietary" labelling -- tells me that you folks really don't understand what this product is all about. That's great, in my book. Took you a while to figure out Centera, if I remember right.

Atmos exposes APIs that anyone can use. It's pretty open as far as its ability to incorporate external policy logic, encryption engines, codecs -- you name it. Pretty open, as far as such things go.

Certainly more open than, say, an EVA?

As to the rest of your nit-picking and "I've talked to customers" nonsense, have a good go at it, please. As far as responses from our competitors generally go, that's pretty weak stuff.

We really enjoyed the two posts you made. The first one (yours) missed the point entirely, which caused a considerable amount of amusement here at EMC. The second one (from your NAS guy) is a bit more thoughtful, so we give him a bit more cred. Still doesn't get it, though.

But, let's face it, until HP has some sort of offering for cloud-optimized storage (policy engine, etc.) you're not really in the game -- all you can really do is hurl rocks and petulant taunts.

Same goes for flash, FCoE ... -- well, we're back to our old dialogue, aren't we?

Cheers!

[Mike Dutch]

Hi Chuck, I tend to think that Cloud does solve problems in today's traditional data center. COS opens up some interesting ways to simplify BURA/DR and data migration (think "stretch VMotion"). I like how www.active-circle.com can include tape as a tier and though it's NAS/CIFS/FTP-based, I wouldn't be surprised if a little engineering couldn't add an object-interface. As you say, it's all about the use case you're trying to address.

[Dave Tauzell]

Is this similar to Google's GFS or Amazon's S3? It seems to have some of the pieces and then some more (the ability to move data to where it is being heavily accessed, for example).

[Robert Ingram]

Chuck, congratulations with the general availability of your product.

With great interest I have read this new post. However you haven’t convinced me on your third bullet in “what it’s not” just yet. Perhaps you can help us there a bit more.

Retention control of information wasn’t our driver. We selected Centera because of its substantial lower cost in managing large volumes of “active content” which makes the statement that Atmos’ use cases and reasons being totally different no longer hold. I am confident that we are not unique in why we own Centera.

EMC did introduce a new storage paradigm with Centera combining object addressing, RAIN architecture, self configuring and self-healing. Rightfully so earned a place as CAS beside NAS and SAN. I can’t say from the reading so far that Atmos does the same. Atmos seems to build on the success that the Centera product has with applying these techniques. This begs the question why EMC decided not to leverage the successful and proven Centera to add placement policies and network a bunch of Centera’s. One almost thinks that EMC it self does not understand it’s own Centera product that well to recognize the opportunities it has.

[Chuck Hollis]

Hi Robert -- you bring up some great points, so thanks.

First, I can see your point, which is that -- at scale -- many of the challenges that Centera addresses are the same ones that Atmos is intended to address.

I think we might be in a "glass half full or glass half empty?" situation regarding this topic.

On one hand, it might be persuasively argued that EMC should have simply extended Centera's functionality to include the advanced geographic distribution policies, or related capabilities.

On the other hand, it's also fair to say "different use case, different market" and justify Atmos based on the optimization in the product for the specific tasks at hand.

One important influencer in all of this is "will customers need both at the same time?", and the initial assumption was "highly unlikely".

So far, we haven't found a single customer (yet) that needed both.

Thanks for your thoughts ...

[Paul Warren]

Hi Chuck,

Great blog. Very inmformative.

I notice there is quite a bit of positioning, by Microsoft, of Azure against Atmos and Vmware. They seem to be playing down atmos and positioning azure as a platform as opposed to vmware as infrastructure only.

Will you be commenting on this? Love to hear you thoughts.

_Paul

[Chuck Hollis]

I don't see any overlap whatsoever between what Atmos does and what Amazon's EC3/S3 and Microsoft's Azure might be doing.

The latter two are primarily application environments. They target software developers and very small companies who need a somewhat scalable application delivery platform and usually can't consider traditional IT alternatives. Great, there's a need for that in the world today.

Atmos is quite different: it helps manage mega-content on a mega-scale. Its target is (typically) large organizations who want to be in the business of capturing large content from around the globe, and distributing it around the globe -- in a cost-effective, automated and secure manner.

Neither Amazon's offering nor Microsoft's offering does anything like this. Maybe some day, but certainly not anytime soon.

VMware is another story -- it takes existing applications and wraps them in a virtual container, making them easier to run better in the data center, or in the hands of a service provider (sometimes called cloud, but not in the most accurate sense of that word).

With Amazon's and Microsoft's offering, you're developing a new application using their tools. With VMware's offering, you can run existing apps as well.

There are more differences, but I think that's a key one.

[Charlie McSweeny]

This would be a good solution for internal delivery of the high density video training we are currently trying to squeeze through a 400kbps pipe to the COEs ... many of our engineers would rejoice to have this solution implemented immediately. What's the chance we could do that?

[Chuck Wegrzyn]

I know you are self-promoting Atmost as the end all, but check out twistedstorage.sourceforge.net - it was out in the open source community over 3+ years before you guys even thought about the idea. I ought to know since I met with the biz-dev people and Fred Olivera to discuss storage. They were trying to find out what I knew!

I started talking about "COS" long long before you guys even considered it. I talked to every major VC in the area about where storage was going to move, and this was late 2005/early 2006.

Today my little open source project is in V4 and deployed at sites around the world, and as of today have two very important institutions evaluating it for wide use in their organization!

[Chuck Hollis]

Congratulations on a successful open source project! I'm not entirely surprised that the ideas behind Atmos and COS have been rattling around for a while.

Then again, you'd probably appreciate that not every organization is comfortable using open source directly -- our friends at RedHat and Novell/Suse can attest to that.

Thanks for sharing the background!

-- Chuck

[Sundar]

Are large players like Akamai planning to use Atmos Cloud for edge content delivery and aggregation?

[Andy Sparkes]

Hi,

I've been trying to follow Atmos for a while now and haven't seen much in the press. I'm genuinely interested in the adoption of these cloud storage technologies. The issues that I run into with customers is "lock-in", thats not to be confused with a particular hardware platform but rather the lockin that occurs through capacity and bandwidth - as you store more and more it becomes more and more difficult to move to another cloud source -we all know about special offers to entice new customers but ultimately its the current customer base that subsidises that - with your cellphone you can just move, but with a few 100TB of data stored its actually pretty expense to move it out and then into another provider. Ultimately doesn't this hurt the economic arguement around cloud?

[Chuck Hollis]

Hi Sundar

Akamai's strength (as with all CDN approaches) is pushing content to millions of consumers around the globe. While Atmos *could* be used for that purpose, I think its strengths lie elsewhere.

For example, a CDN approach would be a poor answer to models where content is generated everywhere, and consumed everywhere. Likewise, not a good fit when you want a long-lasting repository of content, rather than today's hot viral video.

Again, both approaches have their strengths and weaknesses.

-- Chuck

[Chuck Hollis]

Hi Andy

You ask an insightful question, since -- undoubtedly -- it's very possible that capacity exceeds bandwidth in many situations.

I think it won't be as bad as you think for two reasons.

First, we're already seeing a few "federated clouds" where people are considering using an Atmos model that is comprised of several infrastructure providers, all under a single domain of control and security. This has the dual effect of (a) spreading the risk around, and (b) keeping everyone honest.

The second thing to consider is that 100TB can be moved over a network, it just takes a long time. So, if you're patient, the move shouldn't be that expensive.

Thanks for question -- it made me think a bit ...

-- Chuck

[Andy Sparkes]

Thanks for responding.

If you look at the historical trends of bandwidth verse capacity there is a 1.5/1 differential which leads to an exponential separation between the two. This means the problem is getting worse rather than better and that 100TB will grow to something that probably can't be moved over a network. The federated approach may help but you still have that control point, in your case Atmos and I think that most users hate control points in their data paths. It maybe that I'm currently working in a space where 100TB's a day being created is not unusual but not representative of most use cases but its probably a forerunner of whats to come. I am trying to overcome this issue of cloud as its seen as a requirement. I don't have a solution for this problem yet but it maybe that perhaps removable storage (Tape, RDX etc) with a logisitics company might be part of the answer but intuitively it feels like a backwards step and its making me seriously question the hosting of large data sets in the cloud.

[Chuck Hollis]

Hi Andy

Good points. I accept it as an article of faith that data repository size will continue to outstrip available bandwidth. That was true in 1985, and likely to be true in 2015.

So, the strategies are (as we both have pointed out):

- slow dribbles
- have the information live in multiple locations
- ship physical media around

The new bandwidth of several large arrays in a truck needs to be calculated! 100TB is less than half a rack of storage these days ...

-- Chuck