I wonder if someone has implemented in his production environment MSCS cluster where 1 node is Physical[/b] and the other is virtual[/b]. (VM resides on ESX 3)
I assume that the cluster LUNs resides on the SAN as NTFS, and not on the VMFS.
The same question is relevant for NLB \[Network load balancing] cluster.
Thanks in advance
correct, sort of, you need to use an RDM ( at least for the shared disk - not vmdk on vmfs ) so that both the virtual and physical can see the shared disk.
Here is a good doc for getting going: http://www.vmware.com/pdf/vi3_301_201_mscs.pdf
The last time I did this was in the 2.x days and it worked well
We do this right now, and I have not any issues. It actually was easier than I expected.
Was this on MSCS W2K/W2K3? And what about NLB? Did you check Vmotion?
Also do you have high level steps for implementing ?
as per the doc linked to above:
"You cannot use VMotion on virtual machines that run cluster software."
The doc also contains high level steps for 3 different configs: cluster in a box, cluster across hosts, and one physical and one virtual cluster configs.
Was there something that was missing that there are questions about?
I can not find the section which describe the high level steps for "physical and one virtual cluster configs" - could you please specify which section in the pdf file is relevant?
Also can be have a "half" cluster on NLB?
I've done this in VI3 successfully. However I can't agree with the "Did not have issues part". For one, assuming you are using any level of SAN redundancy at all, you will have multi-pathing issues with this sort of setup. Officially VMware does not support the Physical Host having more than one path to the backend storage (NOTE in the case of most SANs this means not even a path to both SPA and SPB). This is because VMware multipathing and a lot of other multipathing software are not compatible to the same LUN (EMC PowerPath included).
However, this only really presents a problem when booting. So, for instance, if the VM Passive node is on when the Physical node goes to boot... It will hang. As long as you shutdown the passive node before you boot the Physical one, you are fine. The VMware passive node can boot just fine with the physical node up and running (This was a 4 path system).
Also, as of when I last did this with VI3, the default 4GB drivers were not officially supported for MSCS clustering and you had to switch them out for 2GB drivers.
Beyond that no issues, but I personally strongly recommend just virtualizing both nodes. Its a better, and more supportable, overall solution IMHO.
Page 35 chapter 4.
Thanks. But what about the NLB implementation \[1 Phy & 1 VM]
There is no doc on that that I have come across. If your are using NLB and spreading the load around 2 or more machines I don't think there is much difference between a physical and virtual machine unless the server is meant to run your NLB. If it is just another machine to distribute load across there isn't much to it.
I am referring for example to
Virtual MAC behavior during VMotion, etc..
NLB is much more forgiving than MSCS. NLB does not run on one particular machine. It is pretty basic, there is a VIP outfront and NLB nodes on the back end. Traffic can be load-balanced a number of ways, I think the default is round-robin. The set-up of NLB is the same no matter the mix of physical and virtual machines. The gotcha with NLB is there is no replication of any sort, just need to make sure the content is the same on all your NLB nodes for consistancy's sake.
I have set up multi-node active/active clusters using VM's and have never had an issue. I imagine physical to virutal will work just fine, however, won't be supported by anyone. Not sure that is a good solution for porduction.
Also make sure that you check the SAN compatibility guide.
I found out the hard way that HDS 9500V wasn't certified by VMWare for MSCS. So even though Hitachi said they had tested it and it was "supported", they didn't get it certified to VMWare's standard so when I had a support issue I was on my own.
FYI. As of the 8/7/07 SAN Compat guide. Hitachi support MSCS clusters on most of their HW in VI3.