Hello Community -
Currently on a site doing a SRM to protect Virtual Machines and Platespin Protect (from Portability Suite) protecting physical machines.
All has gone smoothly over the 2 days. Back tomorrow for final day. All is good but one fact.
The customer was advised to purchase 5 SRM (1-CPU) licenses only as there are only 9/10 Virtual Machines which will require fail-over from PROD site to DR site.
This makes IP customisation a simple with the new IP Customisation tool, but SRM licensing tricky.
In the past I've put SRM in 10-12 hosts clusters, licensing covered up to 24-CPUs since the each host was two socket, dual/quad core...simple right.
But this is different.
The Cluster is made up of 12 hosts, a majority are single socket, some dual socket. Total Virtual Machines is around 200. Only 10 of these virtual machines require protection services from SRM.
I have 5 licenses. Each VM has 1 vCPU.
How do I approach this scenario? Is there a way to tie SRM licensing to a single socket on a particular host? Is there a way to specify only some hosts get the protection and other do not?
Will we need to create a separate Cluster? Or maybe use DRS rules to prevent some Virtual Machines from moving between hosts.
The final solution might be to allocate 4-CPU SRM licenses to 2 Hosts with dual-sockets, then pick one of the single socket Hosts and allocate the remaining 1-CPU SRM licenses.
This would give greater coverage than the 10 Virtual Machines, not a problem.
The question then is: How to ensure these Virtual Machines will always live on these 3 ESX Hosts?
there are ways to do this though all have pros/cons.
you are spot on that the simplest headache free way is to simply license all hosts in the cluster or split some hosts off into a smaller cluster and license them.
as you know there is no such thing as a DRS "host affinity" rule....but this may come...stay tuned....so there are some basic options that can allow you to artificially tie VM's to specific hosts in clusters.
1. create some ghost/dummy VM's on all hosts where you don't have enough license coverage (could just be empty VM's with no GuestOS), disable these ghosts from HA and DRS so they stay tied to the host there on and then create DRS affinity rules for all your protected VM's so that they always stay away from the ghosts (how ironic that I am suggesting this so close to Halloween). This will work but is going to be a real pain if/when the customer starts protecting more VM's or adding in new hosts, not flexible or dynamic at all.
2. only present the replicated storage to the hosts you want the protected VM's to live on, this will prevent them being move to the other (non-SRM) hosts.
Now there are obvious major disadvantages in both of the above suggestions as they introduce complexity and management overhead that most customers would gladly not want but they are just some examples. If it was my environment I would go for the cluster split approach as then I can still have all the flexibility I had before without having to worry about DRS affinity settings or lun presentation etc etc.
hope this helps,
Thanks so much Lee.
Just tell me more about what you mean "Split Cluster" approach.
Is there a way to just license per Cluster as my thinking at the moment is that its done via centralized (spelled USA way) vCenter Server.
Or are you alluding to the PROD_SRM and SRM_PROTECTED_HOST distinctions?
sorry I think I should have used your term...split cluster is basically what you said..."separate cluster". Once you have a second cluster containing a smaller subset of hosts then obviously you have enough licenses to cover all sockets in that cluster. As soon as you put a protected VM on a host that host will try and withdraw the equivalent number of sockets licenses for SRM from your license server equal to the number of sockets that hosts contains.
hope that helps,