VMware just announced some cool stuff at VMworld, and the show is just beginning. I will be there in Las Vegas Tuesday through Thursday this week to catch up with friends in the virtualization ecosystem, but it looks like I already missed the big announcements.

It’s a great idea to federate between on premise and off premise clouds, allowing enterprises to use compute capacity on site and then to use off-site clouds when capacity peaks. Some call this cloudbursting.

So why are the enterprises yawning? Right now, this architecture is mostly useful for scientific computing and number crunching, neither of which often sit on VMware in the first place. It’s not that useful for transactional web computing because the vast bulk of transactional computing requires back-end database and storage access. Sending your spike traffic off-site, to a cloud that doesn’t have an up to date copy of your database, isn’t going to be that productive for most types of web transactions.

This capability will become much more useful when apps are actually designed to span multiple clouds, but in those cases the files and database backend will need to be replicated across 2 clouds in near real time. Pricey, but we’ve been doing that since the mid 90’s for high end enterprise sites. It also means that clouds are sticky - if an enterprise is replicating between 2 cloud infrastructure providers, do you think they’ll sign with a 3rd? No way.

But here’s the funny thing: the more servers you have, the less you need VMware. Why? Because you start to think in units of racks instead of individual servers. Why give up 10% or more of your total CPU (the virtualization CPU performance tax) in order to be able to subdivide a single server when you have 1000 servers? My own experience launching an early cloud offering taught me this, and I’ve heard it echoed by operations executives at large web compaines like Google and Facebook.

For an enterprise that has 20 servers, but wants to consolidate to 10, VMware is great. But for the enterprise who has a major Web 2.0 application spanning dozens of servers, VMware’s vCloud is just not interesting for performance reasons, especially compared to a cloud provider willing to reprovision bare-metal servers using a more traditional data center automation toolset, like BladeLogic or Opsware.

So where does this leave vCloud? It feels like vCloud is the new cpanel for hosters. It’s a good business, but nobody has made a fortune in mass market hosting automation tools, although many have tried. I’d like to see how VMware plans to get around this problem.

I am going to VMworld on my own nickel with very little in the way of a set agenda. If you want to chat about this post, cloud computing, or virtualization in general, drop me a line. bitcurrent@asprey.net