IBM Wants Linux' KVM To Compete With VMware

One of IBM's current goals is to "accelerate the maturation of KVM as a world class hypervisor." That may not sound like much to the uninitiated but IBM has picked its targets well in the past. Of course it's now ten years ago that it announced its backing for Linux.Dan Frye, IBM's VP of open systems development, commented on IBM's commitment to help mature KVM during his address to the Linux Collaboration Summit in San Francisco April 14. KVM is the hypervisor first produced by the Israeli company, Qumranet, and added to the Linux kernel in February 2007; Qumranet was acquired by Red Hat in 2008.

After his address, he sat down to talk about IBM's interest in KVM. Not too long ago, I reminded him, IBM was strongly committed to the Xen open source project, which produced the successful Xen hypervisor and the XenSource company to promote it. XenSource was acquired by a close Microsoft partner, Citrix Systems, and now XenServer and Microsoft's Hyper-V support an identical virtual machine file format.

Frye didn't comment on this ancient history. Instead he focused on the virtues of KVM, although at one point he genuflected to IBM's previous whole-hearted support for Xen.

"KVM is the more natural open architecture in the long run… It's part of the Linux kernel and being in the kernel make it simpler to improve. You don't need to update KVM when the kernel is updated," he said.

Then he added, "It would always be nice to get it right the first time."

I personally think IBM and several other major companies, including Sun Microsystems and Oracle, joined forced behind Xen to create an alternative to VMware's ESX Server. That support rapidly upgraded Xen into a commercial product, which became XenSource's XenServer, Sun's xVM, Oracle's Oracle VM, and Virtual Iron's core hypervisor, now also part of Oracle. But open source developers sensed that Xen was a big company open source project and would serve their purposes in the competition with VMware. They gravitated to an alternative, KVM.

KVM didn't invent a scheduler and memory manager for a hypervisor. It took advantage of the ones already in the Linux kernel. The kernel maintainers found it easy to integrate KVM into the kernel, something everyone had stumbled over with Xen, whose own microkernel architecture had to be constantly kept abreast of changes in the kernel.

Now it's a question whether KVM adoption will ever match the commercial products already established. Microsoft is assured some share of the virtualization market with Hyper-V as a feature of Windows Server 2008. VMware continues as the marketplace leader in virtualization. XenServer is collecting it's share of customers but there's still been no serious erosion of VMware's core high end customer group.

Where does KVM fit into this picture, if at all?

Virtualization is part of cloud computing and open source code thrives in the cloud. An estimated 90% of the workloads running on Amazon Web Services' EC2 are running under Linux. Big data centers on the Internet tend to run Linux and open source code because of its reliability, ease of replication and its contribution to controlling expenses. Competition in the cloud begins from an open source base.

Frye noted that IBM is using Red Hat Enterprise Linux and KVM as the basis for its public cloud. "We spent a lot of time working with KVM (Qumranet), Red Hat and others. KVM has been hardened rapidly. We want to be able to virtualize 64 or 128 guests on KVM. We want it to be equivalent to VMware."

In my opinion, IBM's original motive in supporting Xen -- and it continues to support Xen -- was to prevent VMware from running away with the virtualization market. If VMware did so, many of IBM's future engagement with clients would be dependent on the technology of a proprietary company outside its control.

IBM invented virtualization with the VM operating system on the mainframe in the early 1970s. The idea that its future in x86 server virtualization would be dependent on an outsider was difficult for it to swallow. But plowing its expertise into Xen turned out to be something of a dead end when that technology ended up in the hands of Citrix Systems. That had never been the game plan either. Who could have foreseen that a major open source project would end up speeding Microsoft's entry into the virtualization market? Now IBM had two proprietary competitors instead of one and little influence over x86 server virtualization.

So now IBM is at work a second time, trying to "get it right." It's sticking closer to the roots of open source code where many contributors build a project. As a major contributor, it's already close to the Linux kernel development process, where kernel upgrades that include KVM roll out on a regular schedule. And Red Hat as the owner of KVM is a committed open source company.

IBM continues to support Xen, but its development resources have moved behind KVM. And KVM may yet prove to have a major role to play in cloud formation.

Perhaps more significant than IBM's choice of it for the IBM test and dev cloud is the fact that NTT Communications in Japan has built its BizHosting Basic cloud service on Red Hat Enterprise Linux and KVM virtualization. Not many companies, in either North America or Japan, are necessarily running KVM in the data center. But it's easier to build migration and conversion tools around open source code than proprietary code.

A second tier of cloud providers is coming along, such as NTT Communications and The Planet, and they will learn from the experience of the pioneers. Many of them are either deciding to support KVM or use it as the core of their service offerings. The Planet runs seven co-location data centers with 20,000 customers. Eighty percent of them run Linux. Some of them will adopt KVM was their cloud workload hypervisor, capable of rotating jobs between on-premises and in the cloud.

Frye in his address to the assembled Linux developers at the Collaboration Summit offered a few choice words reflecting IBM's experience with Linux, Xen and KVM.

He noted some major IBM code contributions to Linux that were not accepted. He noted how IBMers needed to learn how to participate in a code contribution process and not drop off a lump sum of code and expect it to be adopted. "No one controls the open source process," he said. "If you try, you make things worse."

IBM is now riding with the KVM development process, trying to toughen up the viability of the hypervisor with its expertise and seek new engagements based on its adoption. Frye knows he's not in control of the process but KVM has now entered the lists and cloud computing will be the next battleground of virtualization adoption. IBM will try to see KVM is as well equipped as possible for that fight.

InformationWeek has published a report on virtualization management. Download the report here (registration required).