There are a growing number of open source cloud infrastructure players in the market, but Eucalyptus says it is the best bet for private enterprise clouds. Network World Editor in Chief John Dix caught up with CEO Mårten Mickos, who, as the former CEO of MySQL, has loads of experience growing significant open source companies, to learn more about the company and how Mickos sees private clouds evolving. Mickos addresses where Eucalyptus fits in, typical use cases, how they are different from Piston Cloud and OpenStack, ties to Amazon (described as the Linux of today), the importance of migrating jobs to and from public clouds, plans for the $30 million in capital Eucalyptus just raised, and more in this wide-ranging chat.
You describe yourselves as an infrastructure-as-a-service software company, but why not just call yourself a private cloud software provider?
Many times we do call ourselves a private cloud software company. I think from your perspective, when you look at us and other vendors in this space, we all have different terminology for what we are. Others call themselves an orchestration layer or a cloud operating system or a cloud platform or private cloud platform, and they all essentially mean the same thing, which I think shows this market is still in its formative stages.
We see our role as bringing the power of cloud into the enterprise. On your own servers, behind your firewall.
When you need the power of cloud inside the firewall under your own control, we bring that to you, and we do it specifically on the IaaS layer, so we are not trying to be a PaaS vendor or a SaaS vendor. We stick to the infrastructure layer, deep down in the plumbing. What we bring that others don’t is we’re an open source platform, we follow the industry standard API on this level, which allows you to move workloads freely between the public cloud and the private cloud. And we have technology and features that nobody else has produced, most specifically, high availability of the Eucalyptus service itself. So we are the first private cloud software platform to provide HA for that very cloud that you run on Eucalyptus.
Our main competitor is vCloud Director (VCD) from VMware. And they come in saying, “Hey guys, you already have our hypervisor and we can add cloud as a feature on top of the virtualization layer.” That’s the essence of their message.
IN THE NEWS: Eucalyptus makes move to stand apart from open source cloud rivals
Are there various components to your product?
It is just one product and you get all of it. You get the Eucalyptus platform and it orchestrates your compute, your storage and your networks all in one. And that’s actually a differentiator in respect to some other products. It’s QA’d. When we produced our latest release, 3.0, we tested it over 15,000 times in house before we shipped it. So we pay a lot of attention on testing the full product, and we don’t ship it as different projects or different components. It’s one product that behaves the same and is consistent across the various areas of functionality.
So it’s server software, but how do you control the storage and networking?
Our role is to come in and land on whatever infrastructure you have. So you say, “I have EMC, I have NetApp, I have this, I have that,” whatever you have, and our product is capable of interacting with all of that. As you configure Eucalyptus for your environment, you specify whatever you have in terms of compute, networking and storage.
Is there a sweet spot in terms of deployment environment? Is it ideal for X number of servers with Y type of networking and Z type of storage?
Size-wise there isn’t a specific sweet spot. People get going with very small clouds, just a handful of servers, or you could have thousands of them. So I would define the sweet spot in terms of use case. And the No. 1 use case is a scalable Web service. Take Puma.com. All their consumer-facing websites run on Eucalyptus because they needed that sort of scalability and elasticity to serve the various campaigns and mini websites they run, so it’s a perfect fit for us. Similarly, Intercontinental Hotels, gaming companies such as Plinga, Riot Entertainment, Electronic Arts, all of those are running scalable Web services that can also include a mobile component.
Can you provide more details in the case of, say, Puma?
I am not at liberty to mention how many servers Puma.com has, but it is a high number. What they get is the ability to shift workloads over to wherever they need it. So as they launch a new website and they see how traffic is growing, they can spin up more virtual machines to serve it. And they know the overall workload is reasonably stable so their data center can handle it. But within those various websites, they have varying and unpredictable workloads and they monitor that and they spin up more resources where they need it for the time being. So they get a performance benefit because, compared to other solutions they tried, this gives them more performance.
So scalable Web services is one common use case. Are there others?
The next one is dev and test environments. So larger organizations with development teams may need to spin up the environment to test a feature and then they don’t need it for a long time. So they are transient application loads. By powering it with the cloud you can have the same physical servers serve all of the developers. Whereas in legacy environments you see every developer has a server under his desk. Well now you can test stuff whenever you need to and you have the whole cloud at your disposal. So it speeds up development. And that’s particularly important today where you test many more variations. You must test for different mobile environments, different operating systems, different features. And now you can test a single feature. You add one feature to the program, spin up a test environment with tens or hundreds of instances, test it, and then go back to development. So a much faster, sort of Darwinian cycle there, with trial and error and figuring out what’s wrong and what’s working. So it makes the development theme more productive.
The third typical use case is big data and scientific applications. So areas where you have a lot of computation or a lot of data that moves back and forth. And that’s actually a reason for doing it in a private cloud, because that huge amount of data would be expensive and slow to move over to a public cloud, so you choose to run it on your own servers because you gain so much in terms of bandwidth and latency and performance.
So I take it customers bring you in to solve specific problems, instead of saying, “Hey, come solve my server utilization problem across the organization.” It’s still more siloed for specific uses?
Very true, and the word “still” is right, meaning I see this as being a perfect example of what Geoffrey Moore described in “Crossing the Chasm.” We are still in the early days, dealing with the bold visionaries. But we can see it coming, where a company will go completely over to cloud to improve the manageability of their infrastructure. But for now it is customers are looking for strategic benefits, agility in one form or the other. Slightly longer-term it will be manageability of the cloud, and really long-term it will be economy, in the sense that if you install private cloud software you can increase your utilization. So I think we’ll see that over the coming years, it’s just that it takes time for a market and an industry to make such a shift.
Once customers have put in some of your stuff to help scale their Web services, will they be able to add applications easily enough to build that out?
Absolutely. And we see a lot of what we call “internal hybrids,” meaning a customer runs modern applications on a Eucalyptus cloud on-premise, but they run old applications in an old environment, or they may run their databases as physically provisioned because they conclude that databases have more performance when you provision them physically and the database takes up the whole physical server anyhow, so you don’t get the elasticity benefit there. So we see combinations like that, where not everything can be moved or it doesn’t make sense to move everything to the private cloud yet.
But over time, we believe that every data center will be managed as a private cloud and that’s why we think this is a huge shift in the industry in terms of software architecture and also a massive business opportunity, because at some point everybody will need this.
What kind of time frame are you envisioning?
How long did it take for the world to switch over to Web architectures? That was a 10-year cycle. How long did it take for the relational databases to really break through and become the dominant design? That was a 10-year cycle. Maybe today the world moves faster, so it’s not full a 10-year cycle. Like if you compare it to VMware’s rise. VMware’s technology was around for a long time, but once they started really breaking into the enterprise it didn’t take them more than six years to become a dominant design and for virtualization to be a standard feature of any data center. So I would therefore guess it’s something like six years, eight years.
You know, if you look at the Eucalyptus history, the project started in 2007 as an advanced research project at UC Santa Barbara. So we have now five years of history here and about a little bit over two years as a commercial entity, but five years as thinking of the technical challenges. So that’s a typical time for getting into the mainstream and then it takes another few years to get everything right and turn it into a massive thing.
I asked about the network and the storage requirements, but I didn’t ask about the hypervisor. Will you work with any type of hypervisor?
Yes, absolutely. We’re not trying to sell you a hypervisor, an operating system, hardware, or anything. We are focused on this category and we innovate here and we make sure we can sit on top of any infrastructure you have, including any hypervisor. And that’s a huge benefit for many who run large data centers and need to avoid lock-in. And the hypervisor has become a point of lock-in in the industry. With KVM and Xen really making inroads into the enterprise, that’s good news for customers, and our platform is one that can support all of those and even abstract them away so much that the application won’t know whether it runs on VMware’s hypervisor or KVM or Xen.
I was talking to the CEO of Piston Cloud recently and he said when he worked at NASA he tried to use Eucalyptus to build Nebula but gave up on it. Are you familiar with that history?
I am, yes. If you are going to start a competitor, you would probably say something like that.
But I don’t think he had that in mind at the time, right?
I would ask him. I would also ask whether NASA continues to use Eucalyptus, and the answer is yes. And I would ask them whether NASA uses his product, and the answer is no. But, you know, it’s history. I’m not saying we were perfect back then and we could have handled the situation much differently. I wasn’t in the company then, it was a dozen people here, mostly engineers, not really focused on building a business or working closely with customers. So all of that has changed. The situation would be different today.
Piston Cloud is closely aligned to the OpenStack community. What’s your relationship to that effort?
We think it’s a great project. As you know, the team that built the first implementation of the compute stack for OpenStack took the old Eucalyptus version and rewrote it in Python, and then they started building around and on top and changing things. So we feel we even have influenced the project and given it some design ideas and starting points, which we are happy with.
That said, OpenStack is a separate project from ours and we have different goals that drive us in different directions. I’ll give you some examples. No. 1 is OpenStack was established as an effort to mount an attack on Amazon Web Services and be a competitor to it. So OpenStack tries to compete with Amazon and they are driving their own API standard for that purpose.
When we got going, our founders decided to follow the Amazon API and be a natural complement rather than a competitor. So we have a different mindset. We think Amazon is great. We think they are the leader in the market. We think there’s a lot the world can learn from them. We believe that cloud computing is being defined by them to a large extent and fighting that wouldn’t make sense. So that’s a difference.
Another difference is that we come from the cloud perspective. We bring cloud thinking into the data center and the cloud thinking is very application centric. So we develop our product for the application developers and to make applications free to run on whatever infrastructure they need to run on. The OpenStack project, being started by Rackspace, has a clear operations flavor and sys admin flavor, because it’s a service provider-led initiative. They make the decisions out of that context and it ends up being a different product and with different design decisions.
You can take such a concrete thing as the development language. We built and we are building Eucalyptus on Java because we think Java is great for applications and Java is great for mission-critical, real-time distributed systems like this. OpenStack chose the Python language, which is a favored language among sys admins and which allows for easy changes and modifications. So if you are building a set of projects where you want each end user to customize for themselves, then Python is a good choice. If you, like us, are building a packaged, tested, QA’d product that has an upgrade path and where you go back to the development team to get support for it, then Java makes sense. And technically they may both be as good as languages for producing this product, but it shows a difference in approach. And I think five years from now when we look back, we will continue to see that for these reasons OpenStack and Eucalyptus are inherently different projects.
OpenStack does seem to have a heck of a lot of momentum going at the moment.
They do. They absolutely do, which is impressive and that’s great for them. When you look at how they organize it, you see that it’s service providers and it’s hardware vendors. It’s HP, Dell, Cisco, IBM. And so it is becoming a service provider and hardware vendor-led project. The way we look at it, is we believe innovation in the cloud space is happening in the software layer, so Eucalyptus has over 200 partners, most of them are innovative software companies who are producing cloud services or cloud software because we believe that’s the defining place for the cloud ecosystem. And again, this is a difference in philosophy.
Do you constrain your market by attaching yourself so closely to Amazon?
I think we expand it. I don’t know what their market share of the public cloud is, but it’s got to be 80% or 90%. It’s enormous. And we believe that it is becoming the defining ecosystem. Back in the ’80s vendors started building industry-standard PCs, which meant PCs built the way IBM built theirs. IBM was the defining standard. Then in the late ’90s and early 2000s, open source was the defining standard and Linux was the place where everything happened and the LAMP stack dominated and still dominates. So that’s where you had to be to be part of the action and to reach developers.
We believe that Amazon Web Services plays that role today. Amazon is the de facto industry standard. You could say that AWS is the Linux of today. That’s where all the developers go. That’s where all the young kids learn to program today. That’s what all the software products are built against. And then, as we follow the exact same API, anything that happens in that ecosystem is instantly applicable to us and our ecosystem. So we have absolutely no regrets there, we think it expands our business.
That said, our product is so designed that we can support multiple APIs and whenever there is a second API that starts to measure up against Amazon’s, we will happily support it. We just don’t know yet which one it will be. Will it be VMware, will it be OpenStack, will it be something else? It’s just not known today. Maybe Google will come out with something. I mean I’m speculating now, I don’t know. But when that day comes, Eucalyptus will support the API.
In March you announced developments that would make it easier to migrate jobs from a private cloud to a public cloud. How much demand are you seeing for that yet?
It is an absolute requirement in most of our dealings, but that doesn’t mean that everybody would be doing it. But customers know they must have option value in their roadmap. They all know that the private cloud makes sense only if it can connect to a public cloud. We have a major telecom equipment manufacturer that runs a huge dev and test environment. All their developers are on Eucalyptus and a key reason for them having come to us early on was the fact we have an API that allows them to move workloads out to the public cloud. They haven’t done so yet, but they see it coming and it helped them in making that decision. And this was even before we had a partnership with Amazon. But some customers are hybridized already.
Take Plinga, the leading social gaming site in Europe. When they launch a game, they launch it on AWS. When the workload settles, they move it back into the Eucalyptus cloud without changes, and run the main life span of the game on Eucalyptus. Then when it’s fading away, they move it back out to the Amazon cloud. So there you see a very active hybrid solution, which gives them this ability to move back and forth, but also gives them optionality. If their own cloud is down, they can be completely in the public cloud. If the public cloud is down, they have their own cloud to run everything on. So they have safety in the number of places where they can run those games.
So you’re not talking about bursting, right?
When people hear about hybrid cloud they usually imagine cloud bursting because it’s easy to see the appeal of that. What people forget is that cloud bursting is very difficult to produce. If you have a sudden spike which is nine times your normal value, sure, you can be prepared and go to the cloud, but you must know how to move your data over because most applications have a database. If you have a massive database, you can’t burst it in a second over to the public cloud. You must have a copy lying there already, ready to receive the new application load. To do that you need to keep the database synched with the main one.
When we talk about hybrid cloud, we see more than 20 different usage scenarios and only one of the them is cloud bursting. And maybe one day we’ll have fantastic cloud bursting with the push of a button and everything works well. But there are many other scenarios that make perfect sense long before that. You have disaster recovery, backup, develop on a private cloud/deploy on a public cloud, etc.
How many Eucalyptus clouds are there at this point?
In a 12-month period we detect about 25,000 new Eucalyptus clouds starting up all over the world. So it’s not a cumulative number, it’s an ongoing number. When people download the product, after installation they can optionally click on a link to check for new images. If they do so we see that ping and can count the number of pings. So we don’t see all the clouds, there are more clouds than we can see, but we see 25,000 per year. Now, many of those are experiments and won’t live for long and some of them are tests and some go into production. So how many we really have in ongoing production use, we honestly don’t know. And it’s amazing to look at the geographic distribution because Thomas Friedman was right. The world is flat when it comes to software. Beijing is the No. 1 city for these cloud starts. Others include Tokyo, Taipei, Chennai, Bangalore, London, Singapore, Shanghai, Bangkok, Seoul, Sao Paulo, New Delhi, San Jose.
We’re pretty far down the list.
Yeah, 13. It’s a sign of development. The U.S. is more spread-out. Those Asian cities are more centralized. In the U.S. you have people spinning out clouds in every city, which means that not a single one is really ranking high on the list. So it’s not bad news for the U.S.
How do you guys make money?
MM: We have the same business model as Red Hat, so we sell annual subscriptions. We estimate that for every 100 users we have one paying customer. So you choose as a customer whether you do it on your own or whether you come to us for the subscription which gives you technical support, some add-ons you otherwise don’t get, priority with bug fixes and so on. So business is growing very, very rapidly for us, but it is an open source business model where you choose in which camp you belong. Like if you take the USDA, which we feature on our website, they are a non-paying user. They never contacted us when they built their cloud. It’s an amazing application, but they did it all on their own without ever letting us know until they went live with it, which is a great example of how skilled and determined people can get stuff built on their own for a very low price. They repurposed their own hardware because we support a lot of hardware configurations, so they didn’t even have to buy any new servers to get going.
The bastards. You didn’t get a dime out of them?
[Laughs.] Well no, but they’re very useful to us because now we get great product feedback from them. They are presenting at our user meetings, they are amazing. They’re our marketing department. We’re very, very thankful. And if we talk open source business models, this is why open source works. If you’re a closed source company and you give away free trials to your product, you never get the benefit from the customer until they pay. But we get the benefit from these users even if they never pay because they find bugs, they fix bugs, they build add-on tools and all kinds of extensions. They blog about it, they speak about it, they research it. We just saw Indiana University — they have an advanced research center that is running huge Eucalyptus clouds — testing us against the other open source products, and we came out with flying colors with the best scalability. And that’s priceless.
What was the other open source products?
They had OpenStack there and Nimbus, which is a less known open source product.
You guys just raised another $30 million in capital, I understand, but what will you use that for?
Expanding the business. We’ve opened offices in China, India and the European Union and got instant positive response from customers, and we are using some of the money to expand those presences. But we’re also building out technical support, building out our engineering support, building up professional services and the sales force. We didn’t need the money yet. We still have plenty of the previous round untouched, but it never hurts to be well equipped when it comes to financial needs.
I would presume you’re not profitable yet, but when does profitability start to matter?
I would even say that if we were profitable now something would be wrong. Meaning every time when we close a new deal, our appetite grows and we expand more. And if you don’t do that you can’t be a major player in a new industry of this magnitude. And this market is not big yet. It’s maybe $100 million or $200 million dollars in total, but it will be a multibillion dollar market, and you must relentlessly grow into the new opportunities and that’s what we can do with the funding.
How many employees do you have today?
Eighty. We’re hiring like crazy. I would estimate that we will double every year in terms of staffing. Our business grows faster than our headcount, but still the headcount needs to grow because, even though we have a production-ready product that serves really heavy workloads today, when you go into new use cases you will need new features. And that’s why we will keep developing the product, even though right now there’s massive production loads on Eucalyptus. There will be an ongoing engineering effort for the next several years.
What do you make of Citrix pulling out of OpenStack and moving its CloudStack tech (which it came by with the acquisition of Cloud.com) over to the Apache Software Foundation?
I think that was a bold move and a good one. The CloudStack product is a good product to start with. The group that built it was a skilled team that had their act together. We think Citrix is a well managed company. So all that bodes well for the industry. That said, it’s not clear exactly how it will pan out as an Apache project. Apache has great governance models, but when you go in there you start as an incubator project and then you need to rise in the ranks to become a main project. So I hope that they will keep investing in it to make sure they get there. I’m hoping for their sake they will get more industry support for what they are doing. I think they are a worthy player in this space.
So how do you size up the alternatives out there now?
We see four other main open source cloud platforms. There’s OpenNebula in Spain, which not everybody sees as a main player, but they have very good technology, so I count them in. Then you have OpenNebula in Europe, which has a research-y approach. Many of the customers are research institutions and they’re very strong there, that’s their sweet spot. And then you have CloudStack and OpenStack, which have the service provider mindset. They sell their technology to service providers who customize them heavily. And we build a ready-made product, that has the Amazon API, and we make sure that when people contribute code those contributions stay open. So I see us having different focus in the industry.
You had a major new release last August and followed that up with a new release this month (June 2012). Give us a thumbnail review.
The 3.0 release last year was a massive new version for us. We more than doubled the lines of code. We added many features. We fixed many things. And like Indiana University said — it’s much, much more scalable and robust than the previous ones, and we weren’t ashamed of the previous ones, but it was a major step up.
Feature-wise there were a couple of critical things. No. 1 is high availability, because we have HA of the product itself. Other cloud platforms can run external HA systems, but we have built-in HA that monitors Eucalyptus itself when it is in operation, and it does so on a service level, not on the cloud level, meaning if your cluster controller goes down or if Walrus goes down, it will spin up the hot standby for that and make sure that the Eucalyptus cloud keeps running, whatever happens in your configuration. So that’s a massive undertaking to produce an HA that operates on that level of granularity, and it took a long time for us but we are very happy to have it done now. And nobody else has done that yet. There’s no comparable functionality in VMware’s products or in any of the other open source products.
No. 2 is we added identity management, we implemented Amazon’s Identity Management Protocol. But given our product is used inside data centers, we also added integration with LDAP and with Active Directory. So here you see how a vendor like Eucalyptus will make sure we are fully compatible with the public cloud, but will also implement functionality that is needed specifically in an on-premise environment.
The 3.1 release that just came out joins together the enterprise and open source editions of Eucalyptus — so there is now just one Eucalyptus platform, which is free and open source code. It also has a few add-on integration modules available only through subscription. The most important changes to note are some organizational shifts. For instance, we broke out the QA team to form a separate, focused group that runs hundreds of tests thousands of times to find and fix any and all weaknesses in the software. Our team has spun up over half a million instances testing Eucalyptus 3 to date.
We also have new community projects, like Eutester, which gives our community and customers the opportunity to write their own test scripts for whatever is important to them. And lastly, we’ve created a sustaining engineering team to keep working on early releases and continue to support our heavy-duty production customers.
In terms of new product functionality, Euca 3.1 introduces new features and capabilities, including FastStart, which allows users to get up and running with a Eucalyptus on-premise IaaS cloud in under 20 minutes. Eucalyptus 3.1 also brings compatibility with the latest version of Red Hat Enterprise Linux with EC2, EBS, S3 and IAM using Red Hat Enterprise Virtualization and VMware virtualization platforms. Lastly, with 3.1, all Eucalyptus source code is available on GitHub.
Then we have 3.2 coming out in Q4 this year and for next year we have Eucalyptus 4.0 under design and development.
All right. Well, that was about it from my side. Any closing thoughts?
We didn’t talk much about our partners, but we have more than 200 and they are a major part of the innovation landscape we operate in. And we are very proud of the customers we have running in production. From a competitive point of view, I would urge you to look at how many customers each of those other vendors have. Look at vCloud Director and ask VMware what kind of reference accounts they have. And then you go back to our website and look at all the names we have. I think Eucalyptus stands out as the most robust and most widely used product of all of these.
