Most days I think VMware ESX is the best thing since indoor plumbing. And some days it stinks like the inside of a port-o-let on a hot summer day…
The absolute worst thing you can do on an ESX server is to cause a snapshot to be merged — the massive number of seeks involved bring all I/O on that volume to a crawl. Making the situation worse is that all other management operations (power on / off / reset / etc) will block at “95% complete” until the merge finishes. Being aware of the issue we try our best not to do that, but sometimes it is unavoidable and it sucks to be the person stuck waiting half an hour to power up a VM because someone else deleted a couple of snapshots (Sorry, Chris C).
The blocking issue is just unforgivably lame, we’ll be contacting support about that. And I would be much happier if we could put snapshots on a separate volume — less thrashing plus it would be easier to manage the amount of space that snapshots consume, you’d think that “Enterprise” customers would have already demanded that feature…
