Virtualization Automation – A Proper Virtual World

Frustration with KVM

March 31st, 2009

Virtualization Critical Comparison – Chapter 06

The title of this entry really gives everything away. I have not had a positive experience with my analysis of KVM. I have found KVM to be difficult to install, difficult to configure, difficult to use, and yes, frustrating. At times I enjoy using Linux, but Linux is not easy, not when you want to do something fast and quick. Many will scream that is not so, but objectively, it is true. This is where Microsoft Windows always trumps Linux, ease of install and use. And KVM is no exception. Of course Windows 2008 Core suffers from many of the issues that Linux does, in that Windows 2008 Core it is not trivial, when you have to do things sans GUI.

All I wanted to do was to have one Linux box as my remote management entity (Ubuntu Desktop with Gnome) and two virtualization hosts sans GUI, in effect a Dell 2850 and 2 Dell 2950s, of course NAS via Ubuntu based NFS as the storage back end. This seemed reasonable since I have used this same setup for VMware, Xen, and Hyper-V (using iSCSI) functional comparison in the past. But in reference to KVM, virt-manager and of course libvirt this was impossible on Ubuntu 8.1, since version virt-manager (and libvirt) integrated into Ubuntu 8.1, an unsupported configuration for remote virtual instance creation. Compile of KVM-84, since only KVM-72 is integrated into Ubuntu 8.1 is possible but throws more warnings than the highway patrol about drinking and driving on New Year’s eve. Of course there are 100s of examples of how to do all this via Google and every single one is different, and none of them worked quite right. No disrespect the respective authors; I am sure their respective setups worked, but not mine, they did not setup the way I wanted to setup.

To be fair, I did not make things easy; I used Ubuntu, which is not quite as easy to use as RedHat in some ways, and easier in others, a topic outside the scope of this blog. There is something to be said for commercial quality versus true open source, and Gentoo was even harder to get to a realistic functional level, even though it is my favorite distribution most of the time. RedHat of course, is not quite savvy with KVM, either, given that true integration is not until RedHat 5.4, and was once targeted 6.0 if memory serves. I found the communication of when and how KVM will be integrated into RHEL interesting and telling, but is a discussion for a different day. Of course I don’t have a copy of RHEL 6 or even 5.4 at hand, so Ubuntu it was. I have found Ubuntu 8.1 stable and consistent, but 9.0.4 Alpha 6 not quite where I needed it to be now, so that too impacted my experience with KVM so far.

This is not to say that Ubuntu is at fault, no, it is not. The real issue with KVM is founded in two key issues, at least two for me, even after I have a functional environment. First, KVM and Windows seem to have a love and hate relationship. This is a maturity issue if anything, and I am sure KVM will improve Windows support; it really must to take on VMware at any serious level. Of course KVM will support Linux well, that is an obvious winner. The real war between Linux and Windows has yet to happen, but it will, some day. Windows 2008 Core is just the introductive probe Microsoft has staged to target against Linux. Don’t be fooled, if Microsoft could figure out how to kill off Linux, Microsoft would. Second, I am not a fan of virt-manager or libvirt as yet. None of the basic components are as polished as I need them to be to see any rational parallel or competitive advantage that can threaten VMware vSphere. The libvirt process keeps dying on me for some reason, something that I am still trying to isolate. However, back to libvirt and virt-manager design, I am not talking about the greater feature set, that other platforms already have, including transparent migration, e.g. VMotion, but basic virtualization function. Of the three key components, KVM, virt-manager, and libvirt, I think virt-manager really is the worst; it just does not work well for me, at all. Even VMware ESX 2.0.1 would trump KVM, libvirt, and virt-manager as they exist today. How long ago was ESX 2.0.1 viable? About 5 years ago or maybe a bit less?

Serious development of KVM eclipsing QEMU has been what? About 2 years, give or take a bit? I guess I had unrealistic expectations so soon? Or is KVM losing some steam? Will RedHat dominate KVM to such an extent that it will fail the way Citrix has all but killed Xen? I reserve judgment for now. Do I need to wait for KVM, virt-manager, and libvirt to mature, at the same rate as VMware? I think not, I think RedHat and maybe IBM, among others will drive things at a faster pace. But today, right now, KVM is frustrating, due to its limitations. I really do want KVM to be successful, if for no other reason, than it will keep VMware on its toes! Something that Microsoft has not quite managed to do right so far, did I hear someone whisper System Center Virtual Machine Manager? Since VMware can be seen as the Neiman Marcus of virtualization, what does that make the KVM, virt-manager and libvirt solution, the Wal-Mart of virtualization? Not sure, I think Microsoft is really pushing to win the Wal-Mart parallel in the virtualization market. Therefore KVM will have to define its self somewhere between Wal-Mart and Neiman Marcus parallels?

Entry Filed under: A Proper Virtual World