Citrix Systems’ Simon Crosby: Xen and the Art of Cloud Computing

In his keynote speech at LinuxWorld Expo in San Francisco, Simon Crosby, Chief Technology Officer for Citrix Systems, announced the company’s plan to open source Project Kensho, a set of tools that will help migrate virtual machines between different hypervisors and virtualization platforms.

I sat down with Crosby, who joined Citrix from XenSource, where he had been chief technology officer prior to its acquisition by Citrix. We discussed the role of open source software in virtualization, how virtualization contributes to cloud computing and the role of an unlikely partner — Microsoft — in the development of the open source hypervisor, Xen.

Listen to a podcast of the interview (14:53 minutes).

LinuxInsider: Xen began as a research project at Cambridge University, and then it was spun out into XenSource. In 2007, Citrix Systems bought XenSource and released Xen to the open source community. What is the rationale behind maintaining Xen as open source software?

Actually, there’s an ordering problem there. Xen was always open source, even from its research days at University of Cambridge. XenSource was started because some large users were asking us to support the code base. And so even though there’s a productive community around it, some of the large users wanted a company based around it to take the technology forward.

And so it’s always been self-evident that the most powerful way to develop software is to have a large open source project where all of the key stakeholders from a business perspective can contribute to optimizing the code base for their own needs. That way you end up with the broadest possible influence and collaboration.

And so Xen has moved forward tremendously fast as a result; it supports more platforms, more CPUs, it scales better, and it’s once again validated this notion that open source is by far the best way to create interesting, complex software systems.

LI: How does that benefit Citrix in a business sense?

Enormously. First of all, Citrix is going into an adjacent market. So Citrix’s core business today is based around application delivery — it’s primarily know for a product formerly called “Presentation Server,” now known as “XenApp,” which is about delivering Windows applications to end users. An adjacent market is application delivery and data centers generally, and Xen is a fundamental enabler of application delivery in a dynamic data center environment, because it’s the de-coupler.

And so Citrix’s acquisition of XenSource allowed Citrix to get into the server virtualization business, getting into the data center business. But there is another directly adjacent business to our core XenApp market, which is delivering desktops as a service. That category is known as “VDI,” virtual desktop infrastructure. We think of it as just desktop delivery, and XenServer is a fundamental component of that stack. Completely integrating within that stack gives us tremendous performance, scalability and pricing advantage in delivering our solutions to customers.

LI: Xen can be found in many devices, even handhelds. What value does it bring to smaller devices?

In the PDA environment — and this is a prototype still — there is now an active ARM development group in the Xen project. The arguments are to reduce the number of CPUs in a PDA.

Typically, a PDA has two or three CPUs, mostly for reasons of security, and the code that runs the wireline, which is facing the carrier network, is separated from the base software that is installed on the device when it’s shipped to you, which is again separated from additional software that you could add to the system. Mostly for security reasons, to prevent some virus bringing down the carrier network.

But virtualization gives you this great way of partitioning software securely, and so if you can reduce the number of CPUs, you can increase battery life and you still have all the benefits of security.

LI: Xen is an open reference standard hypervisor. What does that mean, exatctly, and how does it differentiate Xen from other hypervisors?

What I mean by that is that I’m not bothered at all by alternative implementations of the architecture. Microsoft Hyper-V hypervisor, an implementation of that architecture, it’s approximately free — they charge (US)$28 for it — we make it compatible.

Our agenda has been to have ubiquitous distribution of fast, free, compatible virtualization. And to do that, you have to be manically passionate about the interfaces where your software — your hypervisor — meets other software. So that’s the interface where hypervisors meet guests — we’ve got to support Windows and Linux, and we’ve got to be able to boot and run anything that comes toward us from anywhere. And we’ve got to support a rich ecosystem of management vendors who want to manage virtualized infrastructure.

So there we work on standards-based interfaces. And so this explicit statement is there is no value in Xen, except it’s profoundly valuable.

It’s no value there because the interfaces are standardized. It will move forward at a rate which is faster than any proprietary hypervisor code base can move forward — faster than Hyper-V, faster than VMware with ESX — but it will always support open standard interfaces to maximize the uptake of virtualization as a component technology.

LI: The open source community makes sport of taking shots at Microsoft. However, Redmond has been involved from very early in Xen’s history, as you mentioned. What role does Microsoft play in the virtualization space, and how cooperative has it been?

We found ourselves in this rather schizophrenic position as a startup based around the open source Xen hypervisor having a deep contractual relationship with Microsoft. Hyper-V’s ability to run operating systems other than Windows in terms of the enlightenment, which is paravirtualization, Xen-style, is due to us. We do that. We did that under contract to Microsoft in a project called “Satori.” Satori is a pun on Zen, it’s a state of enlightenment. And so why would we do this? Well, for a start, we’re not only in competition with Microsoft for service outlets ’cause that thing is free.

So if you don’t care about the value of a thing then you can’t be in competition for it. And so we didn’t have a problem with working with Microsoft in that way — Microsoft particularly wanted the open source operating systems — Linux, and indeed others; Solaris, BSD — to run in an optimized way on Hyper-V. So it turned out to be an extremely deep partnership, a good one, and it remains that way to this day. Citrix has always had a good relationship with Microsoft, and so it’s been very easy for us to continue.

LI: What role does virtualzation play in such initiatives as Amazon’s EC2 cloud computing offering?

It’s the fundamental building block of any virtual, hosted offering. So the basic unit of transaction between me and EC2 is a virtual machine, which I can just drop in to their data center and run. And of course, it has to be secure, we have to be able to reason about how much resources it needs and bill for it and charge for it, and so on. Xen is widely used for that. It is the technology that Amazon uses to build out EC2.

LI: Xen has an interesting way of enforcing its standard — through its trademark policy. Why was this method chosen?

We don’t really enforce it — it’s voluntary, because everything in open source is voluntary. The danger of having an open source code base like Xen, which is tremendously valuable, is that anybody can take it and arbitrarily change some interface, in which case suddenly their implementation of that stack is basically proprietary.

Because it won’t run virtual machines created by someone else for their implementation of the stack. And that of course breaks the promise of virtualization, so one of the challenges that you face in using open source as a vehicle to get to that point is that any vendor can take it and make their world incompatible with the others, and so we get a bunch of posturing by the different Linux distros and various other folks saying that their Xen is better than others — which is just utter nonsense, because it isn’t — it’s Xen.

So the challenge of course in open source is if your business model revolves solely around your brand, Red Hat, say or SLES or whatever it happens to be, and there isn’t any intellectual property in the thing, then you have to imply to the customer that yours is better than the other guy’s, because otherwise there’s no value proposition. And I just think that that’s a broken approach, it needs to change.

The people who are making money out of open source are the people who add value to it. So the Xen project specifically commoditizes the engine, it does not commoditize the car. It explicitly encourages the multiple vendors to build different kinds of cars.

So, embed it in a PDA? Great. I hope somebody’s making a ton of money out of that. Embed it in a server from HP and Dell, we make money out of that. Embed it in Linux or other places? Good. Let other people make money.

It’s down to how the customer wants to acquire the technology, but fighting about whose version of it is better is wrong because what it does is directly conflicts with the customer’s value proposition of virtualization, which is that the software stacks — virtual machines — are liberated and can run anywhere, and that you have forward and backward compatibility.

LI: And so as we said earlier, Citrix gets its value out of it as well?

Indeed, Xen server, our our product, is a mixed source product. All of the work that we do to optimize the performance of Windows in XenServer is proprietary. We build a complete competitor to VMware’s product — we’re not an operating system vendor.

And Xen Server is tremendously successful, we’ve grown very fast — north of 4,000 enterprise customers — and it’s the combination of open and closed source software that actually gave us a business model. As just a company around a bag of bits, we weren’t going to get anywhere around a support model.

You announced today the open sourcing of Project Kensho. What, briefly, is Kensho and how does it contribute to advancing the needs of the IT world?

Kensho is a project which is Citrix’s implementation of the Open Virtual Machine Format (OVF) — now this is a format that was developed collaboratively between ourselves and VMware, Microsoft, IBM, HP and Dell, and we pushed it into the DMTF, which is a standards body, which is standardizing the management of interfaces for virtualization.

And what OVF does is it avoids what was a looming VHS-vs.-Betamax war between different virtual hard disk formats for virtual machines. So VMware with their VMDK format, Microsoft with VHD and Amazon uses something called “AMI.” There are others out there. And again that’s a case of incompatibility, which breaks the customer’s value proposition for virtualization.

I want to take a virtual machine and move it seamlessly across all of these systems, including into the cloud. So OVF is a packaging format which is vendor-neutral, and allows virtual machines or sets of virtual machines, packages in this format to be installed on any platform.

Today Kensho implements support for Microsoft Hyper-V, Xen Server and even VMware. We’re working on support for AMI, the Amazon image format, and we’re going to open source this, again because we do not specifically believe that there is any proprietary value in the hypervisor to guest interfaces and we think that that should be broadly disseminated.

LI: Describe your vision of the future of enterprise IT infrastructure, and the role of Xen in that future.

Xen has already achieved a tremendous amount in the time since the founding of the Xen project or even XenSource, we’ve driven the price of a hypervisor down from $6,000 a server to zero, and that’s a tremendous achievement. So the objective is to get virtualization free, at which point we can simply assume that it’s a property of every server. Sometimes it’ll be a property of an OS.

And if it’s just a property of the stack, then we can start to build all the cool stuff that virtualization enables, which is a dynamic data center infrastructure generally. So Xen has played a fundamental role in doing that. It’s a seminal technology there. Once you decouple your head from a value proposition around selling hypervisors and start to build a tremendous value for end users, you find in fact that hypervisors are of no value to customers. What they want is the availability of an agile infrastructue.

And that’s the big challenge for IT now, and that’s the opportunity for VIT vendors. So Xen is actually creating a vast opportunity for all of the enterprise IT vendors to meet customer requirements.

LI: In your keynote, you mentioned a project that the New York Times recently undertook as a way of illustrating the potential of cloud computing. Can you tell our readers what that was?

I stumbled on this when I was researching EC2 and the major users of EC2. It turned out that The New York Times had scanned into .TIFFs every version of The New York Times since its first publication back in the late 1800s. This (is) vast — 3 or 4 terabytes of these tiff images.

So what they did was they created a system that ran for a total of 24 hours on EC2, which is Xen-based. One hundred virtual machines using the Hadoop file system, which is an open source implementation of the Google file system. To turn those .TIFFs into .PDFs, then to run OCR (optical character recognition) on the PDFs to allow them to arbitrarily index the data that’s in them so it’s entirely searchable.

And then they created this Web site which now allows you to search any historical version of the New York Times down to any day on any publication ever, including all of the advertisements, and it will instantly bring up the graphics as well as the text, and the text is real text — you can cut and paste and everything else. This was done over 24 hours on EC2, vast amounts of data and compute, and it cost them $240 to do.