Ian Pratt Introduces the birth of, what would become Xen
Xen's history begins with the XenoServers project, started in 1999 at the University of Cambridge Computer Laboratory. Dr Ian Pratt and a number of research students proposed creating an infrastructure of servers across the Internet where any suitably authenticated individual or organization could purchase resources on those machines to perform computation or run services, and do so in a dynamic fashion, ramping up and down their resource usage as required. The project was called XenoServers from the Ancient Greek "xenos" - a stranger to whom you might provide food and shelter but not necessarily trust. The first publication was at ACM HotOS in 1999, "Xenoservers: Accountable Execution of Untrusted Programs."
A more powerful platform
Initially the project looked at using Java virtual machines to run code supplied by customers, but it was soon realized that the platform would be much more powerful if it were possible to run arbitrary existing applications rather than forcing everything to be rewritten in Java. Using a hypervisor and encapsulating applications and their chosen operating systems in virtual machines would be an ideal way of doing this, but at the time virtualization of the x86 architecture relied on instruction emulation and binary rewriting, and struggled for performance.
Instead, an approach was developed whereby operating systems could be ported ("paravirtualized") to run on a hypervisor, enabling applications to then run unmodified. Keir Fraser was the research student that seized on this idea and made it a reality, building the core hypervisor and a first port of Linux 2.2. The hypervisor was named "Xen," a contraction of the parent project name.
One of the key goals of Xen was to enable virtual machines from different customers to run alongside each other on the same hardware, implementing strong isolation such that one customer couldn't steal or corrupt information belonging to another, and that each customer would receive the full resources they were paying for. We had an illustrative design requirement to be able to sell services to both Coke and Pepsi, and for them to be confident about the arrangement. This was central to Xen's design.
First full open source release
Over the following years, the number of students and research assistants working on Xen and XenoServers grew as more people saw the potential for the work becoming important both academically and in the real world, and commercial research Labs such as Intel Research and HP Labs funded and contributed to the work. A first full open source release of Xen was made in 2003, with the GPLv2 license used for the hypervisor, and a BSD license for in-operating system components so as to facilitate porting of proprietary OSes. Microsoft Research contributed to the project, with Paul Barham leading a prototype port of Windows XP.
Later in 2003 the key research paper "Xen and the Art of Virtualization" was published at the leading ACM conference, SOSP. Momentum continued to grow around Xen, with various ISPs adopting Xen/XenoServers to provide commercial hosting services for customers, taking advantage of the project's compute, networking and storage virtualization.
Great platform for academic research
In early 2004, Christian Limpach contributed a port of the NetBSD operating system to Xen, and then joined the Cambridge-based team where he became a key contributor supporting Keir. Under Pratt's leadership the open source community around Xen continued to grow, with companies like RedHat, SuSE, Sun, IBM and AMD joining the community and contributing, porting their OSes to Xen, or porting Xen to their silicon and hardware.
Early in the project it was realized that some extensions to the x86 architecture could simplify the task of virtualizing it, enabling better performance and reducing the amount of privileged code required and hence reducing the security attack surface. Close cooperation with Intel led to the development of Xen HVM mode, utilizing Intel's VTx extensions. HVM mode avoided the need for some of the paravirtual modifications to OSes, making it easier to run unmodified OSes, but paravirtualization was still used where it could provide performance and attack surface advantages. HVM mode later added support for AMD-V.
As well as its increasing role in commercial products and services, Xen served as a great platform for academic research, both in Cambridge and other Universities and research labs around the world, many ideas from which later found their way into products. For example, the Xen architecture was extended with a concept called driver domains, which deprivileged the code used to control IO devices so that it could be isolated in VMs to facilitate seamless recovery from driver crashes or live software updates (OASIS 2003, Safe Hardware Access with the Xen Virtual Machine Monitor). Steve Hand led work to further disaggregate Xen’s control plane to further enhance security. Xen was used to study live relocation of running VMs between machines to enable load balancing and hardware maintenance (Usenix NSDI 2005, Live Migration of Virtual Machines). It was also used to implement techniques like deterministic replay and checkpointing to enable high-performance hardware fault tolerant computation and advanced “backwards debugging” techniques for distributed systems.
Becoming an important infrastructure component
By mid-2004, Xen was becoming an important infrastructure component. Amazon and Google were both using Xen internally and preparing to launch what we now refer to as Infrastructure as a Service Clouds, bringing the XenoServers model to its full potential. In addition to companies building products and services that used Xen, there were an increasing number of forward-thinking Enterprise companies that saw the potential for Xen to revolutionize their datacentres, enabling them to virtualize their infrastructure, or even outsource their infrastructure to someone else's datacentre.
Members of the Xen team were increasingly getting asked to provide consultancy services to support these initiatives, and spotting an opportunity to create a packaged Enterprise product a company called XenSource was formed by the leading contributors to Xen. Getting funding for the company in the UK and Europe wasn't straightforward as business models around open source software had yet to be fully established. This led to Pratt reconnecting with a former University Colleague, Simon Crosby, and a trip to Silicon Valley where there was strong interest and XenSource's A-round was funded by Kevin Compton (KPCB) and Nick Sturiale (Sevin Rosen).
Simon Crosby takes up the story from 2004
In fall of 2004, I was working at Intel, where I’d been trying to get research funding for Ian’s team, but encountering stiff resistance. It turned out that Intel was considering an investment in VMWare… Ian and I and others on his team (there were 7 cofounders in all) met at a conference in San Francisco and decided to start a company. Moshe Bar, who I didn’t know but who had also been talking to Ian, joined us. Nick Gault, our founding CEO was also there. We had real interest from Morgan Stanley as a potential customer, and there was huge traction on the open source code base.
We made a visit to Sand Hill road to a few investors between Christmas and new year 2004, and the company was born in early 2005, with $8M in funding. At that point all effort was on the open source code base and our community. Xen 2 was out and the community was hard at work on Xen 3. TBH we hadn’t thought much about how XenSource would make money but the initial idea was to provide management tools and to get distros like Red Hat and Suse to include the hypervisor with their Linux products. We couldn’t virtualize Windows yet because Xen is a para-virtualizing hypervisor that required changes in the guest operating systems, and Microsoft wasn’t going to do that!
Year 1 of the company was all about building open source but it became clear that the big distros wanted their piece of the pie: they also announced an intention to provide management tools, so we had competition even from open source. At the end of our first year, as a result of our inability to articulate a clear business plan, we lost Moshe and our founding CEO. Happily we met Peter Levine and Frank Artale through Accel Ventures, and they stepped aboard to provide the business sense we needed: at that point we decided to directly compete with VMWare. We also raised our B round of (as far as recall) $25M. The product was a virtualization platform like ESX, that initially could virtualize Linux only, but when Intel introduced virtualization extensions to the CPU, that allowed us to also virtualize Windows, and we added that option. We then developed a cluster product that in my view was superior to VMWare virtual infrastructure.
Microsoft partners with Xen
Along the way we always had questions about how an open source business could stay afloat when all the code was ... up for grabs. We had two real advantages: first, the complex management software for the hypervisor was written in OCaml- which is a rather obscure but type-safe language perfect for the task. Second, Frank Artale had great connections at Microsoft, who were building a hypervisor for Windows but needed a third party to help with open source and Linux, so that their devs would not be encumbered by open source contamination. We agreed to do the work for them, and a fantastic partnership grew between us that saw Microsoft recommend our product whenever they could, over VMWare. As a competitive hypervisor that was recommended by Microsoft, we had the credibility needed to seriously kick start revenue. Our partnership with Microsoft started in fall 2006.
Citrix acquires XenSource Inc
It seems that Microsoft was not the only player upset with VMWare. Citrix, who had initially partnered with VMWare, was disappointed when VMWare entered the desktop virtualization market, competing with their app virtualization business. In mid 2007, given our partnership with Microsoft for virtualization, Citrix approached us as a potential acquisition. We would be a Microsoft friendly platform for them, matching their partnership for app virtualization. Citrix acquired us that fall.
Meanwhile, Amazon was building its cloud using Xen - starting at the end of 2005. And ultimately, with millions of VMs, they were the single largest user and arguably the vendor that benefited most financially. It is absolutely true to say that Xen built the cloud. We had a productive working relationship with Amazon, though it was all related to open source use, and not commercial. Their ability to virtualize windows was based on their eventual adoption of new Intel platforms with virtualization extensions.
A new standalone business unit
Fast forward to September 2022 when Vista Equity Partners and Evergreen Coast Capital acquired Citrix. In so doing they create new standalone business units from the former Citrix, including XenServer.
Get started with XenServer