Techtarget has an interesting article on VM sprawl issue. The key takeaway is that even now, with VMs accounting for only 1% to 2% of OS instances, people already start experiencing VM sprawl. What will happen in a couple of years when this number is up to, say, 20%?
The article also mentions automation as a potential solution. Unfortunately, automation only helps to an extent, and here is why:
1) Errors. The more distinct instances you have to perform a management operation on, the more likely this operation is to fail. This is due to the fact that all instances were provisioned and configured completely independently from each other and most probably have subtle differences, which nevertheless can make the management operation fail.
2) Time. The more instances you have, the more time it takes to complete the operation. Let’s imagine you need to install 20 Megabytes of updates (or 2 Gigabytes of service pack) into a VM. The fastest the update can be completed is the same speed as it does in a non-virtualized environment, and it will need to monopolize the hardware for the duration of the update. VMs compete for the same set of hardware resources, so all other VMs will be stalled. Alternatively, the update can be run at low priority, but then it will take much longer to finish. Now, all of the above needs to be repeated as many times as you have VMs on the machine.
In my opinion, people will find out that automation is not a good enough answer shortly after they face the VM sprawl issue, and there is not that much time left.
As usual, your comments and e-mails are very welcome!