There are a couple of phrases in this article exposing the myths of virtual machine overload which really strike a chord with me. Mike Laverick clearly identifies the increased business risk of trying to load too many VMs on a single physical machine and talks at length about his current anxiety not being about performance, but about availability…

There is some information about sweet spots in memory configurations and the current high cost of 8GB memory dimms which I will leave you to read. All us vendors are suffering in the same way, but prices will come down.

Mike then goes on to talk about the different approaches to providing availability and this is where it starts to get really interesting. He almost dismisses the traditional HA approach simply because VMs and services have to be restarted with no guarantee of what state that VM will be in when it comes up – hence the application is likely to have to go through some sort of consistency-check phase which will dramatically slow recovery time, something I have been saying for ages (read some past blogs)!

There is then praise for the VM FT product, balanced with a paragraph on why the technology is not so cool. I concur. It is very restrictive. That is, of course, why the VMware crew don’t crow about it too much.

What is surprising though is the author’s intimation that this is the first time he has seen continuous availability technology in the real world. I’ve got news! It’s been around for several years in the form of fault tolerant servers. Fault tolerant servers provide ultimate availability. In fact, they don’t fail unless you fill the computer room up with water.

“You don’t need clustering stuff at the virtual layer to avoid overloading your virtual machines. You need a fault tolerant server. Keep it simple, save some costs and increase the quality of service you give back to your business…”

Related Articles