2

As far as I know there is a following limitation for maximum number of nodes in Windows Server 2012 failover clustering: 4,000 VM nodes or 64 physical nodes. Apart from apparent difficulties / management overhead of managing 4000 physical node cluster are there any other reasons for such inequality for max cluster nodes in case of physical nodes? Is it sort of extra incentive from Microsoft to use virtualization?

I guess it is only related with supportability (as extra requirements aplly if you want to get Microsoft support for clusters) and the fact that no customers going to mess with more than 64 physical nodes cluster anyway... Any other ideas / views on this?
Mikhail
  • 1,295
  • 3
  • 19
  • 35

2 Answers2

2

There should be some reasonable limit or somebody would spawn 5000 VMs and crash a GUI.

P.S. 4000 VMs means ~100K IOPS and maybe 200TB of storage. That's a pretty damn cool SAN.

--

As far as I know there is a following limitation for maximum number of nodes in Windows Server 2012 failover clustering: 4,000 VM nodes or 64 physical nodes. Apart from apparent difficulties / management overhead of managing 4000 physical node cluster are there any other reasons for such inequality for max cluster nodes in case of physical nodes?

BaronSamedi1958
  • 13,676
  • 1
  • 21
  • 53
  • (please move to the comments as I don't have enough of the points) Tom that's what MPIO is used for: you run multiple physical paths from a hypervisor node to a set of switches used for SAN traffic (does not matter iSCSI or FC or whatever) so it's a very little correlation between number of hypervisor nodes and probability to have particular node <-> switch data path went South........................ – BaronSamedi1958 Dec 05 '13 at 14:18
1

It is common sense. Given that virtualization is there to consolidate servers in absolutly most cases, what sense would it be for that number not to be very different, especially given that a mid range server these days has like 256gb memory and can run nearly 60 threads.

Also certain scaling issues may appear on the physical node side - remember that this is hwere we talk about CSS (CLuster Shared Storage) and IO redirection. Bottlenecks there will more likely appear than on the VM side. Personally I think that number is high enough - anything higher can easily set up multiple clusters.

TomTom
  • 51,649
  • 7
  • 54
  • 136
  • Tom Hyper-V don't go into redirected mode unless data path is broken (or crazy VSS backup is keeping CSV). Data is sumbitted with data path and metdata is submitted over Ethenret thru the CSV owner. So NO redirected mode in a RL. -- Also certain scaling issues may appear on the physical node side - remember that this is hwere we talk about CSS (CLuster Shared Storage) and IO redirection. Bottlenecks there will more likely appear than on the VM side. Personally I think that number is high enough - anything higher can easily set up multiple clusters. – BaronSamedi1958 Dec 05 '13 at 00:31
  • Yes, and the more nodes you have the more likely it is the path DOES get broken. – TomTom Dec 05 '13 at 07:14