Every major virtualization vendor has spent the last year publishing whitepapers about openness while shipping a virtual disk format nobody else’s software can read properly.
VMware’s format is VMDK. Microsoft’s is VHD. Citrix inherited its own from the XenSource acquisition. Building a virtual machine on one platform and moving it to another is not a supported operation; it is a conversion project, with tools of varying reliability trying to translate one vendor’s assumptions about disk geometry and snapshotting into another’s.
That is a genuinely strange position for an industry to be in eighteen months after everyone agreed virtualization was the future of the data center.
The reason is not technical. Reading and writing a disk image format isn’t hard, and every vendor involved has engineers who could do it in a week if asked. The reason is that the format is the lock-in, and none of these companies particularly wants to make leaving easy while they are still fighting over who gets to be the default.
VMware has the most to lose from openness and is, unsurprisingly, the slowest to move on it. Once a data center has committed real infrastructure and real expertise to ESX, VMware has an incentive to make that commitment expensive to walk away from, and an incompatible disk format is a very effective, entirely deniable way to keep it that way. Nobody at VMware has to say the words vendor lock-in for the effect to hold.
Microsoft is playing the opposite hand for the obvious reason that it is not the incumbent here. Hyper-V is new, VHD is comparatively open already as formats go, and every push toward standardization is a push that makes switching away from VMware slightly less painful, which is exactly the outcome Microsoft wants while it is still building market share.
A standards effort is underway to define a common portable format that any hypervisor could read, and I expect it to happen eventually, because customers with real infrastructure will keep asking for it loudly enough that ignoring the request becomes its own liability. What I do not expect is for it to arrive quickly, because the parties who would benefit most from delay are the ones currently writing the specification.
The practical advice for anyone building serious infrastructure on this technology right now is to assume you are choosing a platform, not a format, and to price the cost of being wrong into the decision. Nobody selling you a hypervisor is going to volunteer that math.