On Monday 07 April 2014 22:52:57 Nico Kadel-Garcia wrote: > Name 2. Seriously.
Rackspace, Cloudera, IBM, Amazon, Google, RagingWire, RedHat, HP, etc. and anyone else on the OpenStack list is deploying it with KVM somewhere that makes money for them. KVM is by no means the *only* thing they use (far from it, and in Google's case only some of the KVM they run is vanilla KVM). That doesn't mean that everything is perfect in libvirt land, of course, but you make it sound like KVM is a dead-end virtualization stack, and its not. Without a decent environment manager (OpenStack/Puppet, for example) its not very scalable on its own. Of course, the price war in cloud provisioning got 60% rougher as of a week ago, so I expect the list of cloud providers running *any* stack will shrink down to two or three in the coming months (Google bleeding Amazon is bleeding everyone else, too, but nobody has as much blood as those two). As for pair-bonding, I can't say I've needed to do that within guests (hosts, sure). I haven't run into a lot of corner cases with KVM. We did with VMWare, but that was before changing our service architecture to degrade gracefully and balance loads better (and, critically, most of the services are now stateless, so bad things are a lot less difficult to deal with), moving to KVM was after this so I can't pin blame on VMWare. We've likely never cooked up a KVM perfect storm -- but I'm curious if you have any insights of what a perfectly awful KVM situation looks like.