Live migration seems like magic - that you can take a running VM and swap its host without the VM or its users really noticing. But in some sense it's not very "cloudy". After all why live-migrate your pets if you can just start a new herd of cattle wherever you need them?
At HP, as operators of a large public cloud, we don't have the luxury of killing all our customers' pets, but for a variety of operational reasons we do need to move workloads around.
In this talk I'll go through some of the reasons we use live migration. I'll cover the problems and challenges we faced and point out some specific areas where we are contributing code back to the OpenStack Nova codebase. I'll also talk through some future use-cases for live migration in datacentre management.
Matthew has worked on Nova and Ironic for HP Helion in Bristol, UK for the last 18 months. Before that he was a closet functional programmer, a J2EE love/hater, a teacher of ESL and maths undergrad.