“Are you planning to use VMware Fault Tolerance when it’s released?”
“Probably not,” I reply.
“Why not? It looks really cool.”
“Sure, if you don’t have stable hardware or a stable hosting environment. There are probably other scenarios that I haven’t thought of where it’ll help, though.”
“What? No… it’ll be cool if you have an application crash or something.”
“Wrong. Fault Tolerance is to VMs what RAID 1 is for data. Whatever happens on disk 0 happens on disk 1. So if you delete a bunch of files they disappear from both disks, and you still need to restore from backup. You only see benefits if one of the drives dies. Fault Tolerance keeps two VMs in sync that same way.”
“Oh, so if my app crashes on one it’ll be crashed on the other one, too?”
“Yeah.”
Interesting, so think of Fault Tolerance as protection from a hardware fault.
Yeah, it looks like it’s exactly that.
This eliminates the need for clustering at the OS level, doesn’t it? Or am I missing something?
Depends heavily on why you’re clustering. If you are doing it to handle hardware unavailability it might be a good solution.
This sounds like the LockStep technology I heard about back in February 2008. I heard about it and instantly thought about using it instead of MS Clustering for our main Database. Thanks Bob for mentioning this, now I have to go and learn about more VMWare fun!! 🙂
I don’t know to me it seems even worse then RAID1, considering some of the requirements I was reading. For example when using NFS the the two VMs need to be on the same share and I think when using VMFS they need to be on the same volume. To me that’s like having RAID1 where your mirroring partitions on a single hard drive. I’m no ESX expert so feel free to enlighten me but how is that any different then VMotion?