Microsoft and XenSource is a one-way street

Brian Byun of VMware talks about the implications of the Microsoft-XenSource link-up for virtual computing
Written by Brian Byun, Contributor

Virtualisation is fast becoming a ubiquitous layer of software that transforms how applications are developed, deployed and managed in IT environments of all sizes. Not long ago, Diane Greene explained the need for three major areas of industry standards around virtualisation that are required to support customer choice as virtualisation adoption becomes pervasive. These standards will enable customers to choose their virtualisation software, management tools and applications, and associated OS on merit-based criteria: quality, functionality and price rather than traditional licensing and arbitrary vendor-specific lock-in rules.

Recently, Microsoft announced that it would work with XenSource to allow para-virtualised versions of Linux to run on future Windows hypervisors. Para-virtualisation is an emerging technology that requires a modified guest operating system to call an API on the underlying hypervisor, in order to optimise the performance of the guest OS when it is run on a virtualisation software layer. In light of this news, it's worth re-visiting some current developments in standardising the interface between the operating system and the hypervisor.

First, let's take note of a few ironies of this recent Microsoft-XenSource arrangement:

  • It's a one-way street that favours Microsoft and Windows running Linux. The arrangement will allow Linux to run on future Microsoft hypervisors through translated calls to the hypervisor when Windows is controlling the hardware, but not the other way around; that is, there is no mention of Longhorn optimisations or "enlightenments" being ported to Xen or licensed to XenSource to enable a Xen hypervisor to run full optimisations with Longhorn OS.
  • XenSource, in diverging from its open source and Linux virtualisation roots, is enabling the commercial interests of Windows and building to proprietary Windows API layers. It stands to reason that in order to protect Windows from GPL contamination, XenSource will need to undertake a lot of non-GPL development to translate and buffer the Linux kernel from Windows hypervisor interfaces; and nothing that Microsoft licenses to, or develops with, XenSource is GPL and can be used directly by the Xen or Linux communities and commercial distributions.
  • In the end, it remains to be seen how relevant this one-way "bridge" to Longhorn will be with Linux customers and the community. After all, no one adopted Linux in order to run it on proprietary Windows interfaces, which is what's contemplated here. In addition, there is a notable absence of support for this scheme from major Linux commercial distributions, OEMs and customers.

Impact of the OS-hypervisor interface
This announcement gives reason to examine the impact of the OS-hypervisor interface more closely. Ideally, any hypervisor should be able to run optimally with any operating system and there should be no proprietary licence required to do so. Under the announced arrangement, XenSource is licensing a Microsoft-owned OS-hypervisor interface, but not the other way around. Microsoft wants to make Windows run optimally only on Microsoft's hypervisor, but is happy to let other operating systems such as Linux run on top of their hypervisor. Clearly Microsoft views their control point as moving to the hypervisor.

We at VMware hope there will soon be a standard Linux interface for para-virtualisation, which will simplify and standardise how Linux is supported on various hypervisors, including VMware and Xen. We are working with the Linux kernel community to develop an open interface so the Linux kernel can run natively and efficiently on a choice of hypervisors. Such an interface would also be available to any operating system. We have made our initial proposal for such an interface available to the Linux community and are pursuing Linux and hypervisor interoperability not as a commercial arrangement, but within the open, transparent and merit-based multi-vendor approach that is the hallmark of the Linux kernel community.

The industry is realising that the x86 virtualisation trend provides a unique point in time where customers can exercise control and freedom of choice of OS, applications, hardware vendors and virtualisation stacks, and where no one vendor dominates and controls. This results in pure value-based competition and opens up a world where any OS or virtualisation approach can potentially run any other OS. It's up to industry participants and customers with buying clout to fully enable this vision. In this new world, we recommend that customers demand unfettered, open standards and full bi-lateral interoperability from all your vendors before expanding or doing business with them.

In the end, I believe most customers will run their Linux distributions on hypervisors that use open standards and do not have licence lock-in to proprietary interfaces.

Brian Byun is vice president of products and alliances at VMware. A version of this article previously appeared on VMware's corporate blog.

Editorial standards