The virtualization track at this week’s Catalyst Europe conference included a lot of really good information from Steve Herrod (VMware), Ian Pratt (Citrix), Etay Bogner (Neocleus), Dan McCall (Virtual Computer), and excellent virtualization deployment and management insight from Alexander Schanz (Head of Datacenter – DFS Deutsche Flugsicherung GmbH).
I plan to post a more thorough assessment of Catalyst EU’s virtualization highlights tomorrow, but I wanted to take a moment to share some really good information. In the track’s final session, I moderated a vendor round table on debating the hypervisor’s future. During the panel, all panelists talked at length on complexities that exist in hardware device models in the virtualization layer and the barriers that differences in hardware abstraction pose to interoperability. If you recall, Citrix recently announced interoperability with Hyper-V, meaning that you could create a VM on Citrix XenServer, copy it to a Hyper-V platform, and run it. That’s it. No conversion necessary.
So if Citrix and Microsoft can get together on this level of interoperability, why can’t Citrix and VMware do something similar together? The answer is that they can. At the panel’s conclusion, Ian Pratt and Steve Herrod agreed to continue the conversation on exchanging device driver models, with the expectation of further simplifying VM packaging by the ISVs that ship virtual machine appliances and also easing hypervisor interoperability. I’m going to go out on a limb and say if when this happens, the entire virtualization community would stand to benefit. Now to be fair, we need at least a three-way exchange at a minimum (Citrix, Microsoft,and VMware), but to hear Citrix and VMware entertain the idea is extremely encouraging. Let me package a virtual machine appliance that I can simply expand and run in any virtual environment, with all necessary device drivers pre-loaded.
Vendors often agree that the barriers we see in virtualization deployment, management, and interoperability are not technical, but political. Political battles often take a long time to resolve, but I’m a patient guy. Let’s resolve them one at a time. Today was a good step in that direction.
Note: Originally posted to Burton Group’s Data Center Strategies blog.
#1 by Etay Bogner - October 23rd, 2008 at 08:02
Just to elaborate a bit more:
As I noted on the panel, the Device Model compatibility problem is the real important one, much more than whether solutions support the VHD, VMDK and/or the OVF formats – which is sort of trivial for me. One can solve it with conversion or better yet, just to support them natively. There are really two formats out there, not counting OVF.
Why? The Device Model problem will grow as more and more pass-through and para-virtualized devices will be assigned to VMs and in cases where a physical-to-virtual conversion has to take place (as Neocleus does) – One does not want to be in a situation where you boot an image and it crashes or becomes inoperable/unmanaged because of a missing driver; at least have the VM boot and have the opportunity to mount your VM tools iso file and install your own set of real and para-virtualized drivers…
#2 by Chris - October 25th, 2008 at 13:10
Etay – you’re absolutely right. Burton Group has been beating the virtualization interoperability drum for about the last two years. Sharing device models is critical and important. Commonality with virtual disk storage would be helpful too, and let’s not forget about OVF. Common XML-based VM metadata for moving VMs between hosting platforms may be trivial to you, but not to most of the industry. VMware and Citrix have full implementations for the OVF spec – far from a marketing checkbox. Moving forward, OVF will be one of the key enablers to cloud-based computing. Sure, it’s not on the radar of Neocleus, so I can see why you would see the standard as trivial. But OVF is an important standard for server virtualization, and will impact hosted virtual desktops in the future as well. At the end of the day, we need interoperability across a number of spectrums, and to your point, shared device models is a crucial piece. I’m hopeful that you will continue to push these issues forward by joining the team at the DMTF and sharing your ideas with your peers in the vendor community who are working to make standards-based virtualization management a practical part of tomorrow’s dynamic data center.
It’s good to see us moving forward for the better of the industry, and thanks again for your thoughtful and thought provoking comments on the panel.