Imagine the following situation...
- A VMware Datacenter with 3 Clusters of ESX hosts; one of those clusters is setup to host all Oracle VMs.
- A CX4 is in production for the shared storage for the three clusters in this datacenter. One Storage Group is defined. All ESX hosts and all luns are in this single storage group.
I want to separate the storage from the two "normal" ESX clusters and the "Oracle cluster". Of course I want to minimalize the downtime of the guests; if possible I do not want any down-time at all.
As far as I know, I cannot add a host to two storage groups, but I can add a lun to multiple Storage Groups.
This is a scenario I can use. I would like from you if I can be successfull this way, and maybe there is another, maybe better, scenario to do it.
- Place one of the ESX-servers in the "Oracle cluster" in Maintenance mode
- Remove this host from the ESX storage group.
- Create a new Storage group "ESX-Oracle" and connect the ESX host to this newly created storage group.
- Connect all luns to this "ESX-Oracle" storage group with the same IDs as in the ESX storage group.
- Rescan storage for the ESX host and check if all luns are visible.
- Exit maintenance mode and put a second ESX host in maintenance mode.
- Remove the host from the normal ESX storage group and add to the Oracle Storage group.
- perform the last two steps for all hosts in the Oracle cluster.
If finished successfully remove all Oracle luns from the default ESX Storage group.
I hope to hear from you.
but how are you separating the storage if you present the same LUNs back to this "new" storage group ? I am assuming you are going to do vMotion at some point ?
I will create new datastores for the Oracle cluster and perform a SVmotion to place the virtual servers at their new datastore... So datastores are shown in two storage groups during migration.
We are also working with Raw Device Mappings. These RDMs will also appear in two storage groups during the migration.
When the migration is finished, cleaning of the ESX storage group will start.
i don't know about RDMs but should work fine with VMFS datastores. So you take node 1 and put into maint. mode, all VMs get vMotion'ed to node 2. You remove node 1 from "general" storage group and place it into dedicated "oracle" storage group. You present the same LUNs to this "oracle" storage group, bring node 1 online and vMotion some VMs back to it. Once you go through this excersise with all nodes, you will present new LUNs and sVMotion over. When that's done you will "release" the old LUNs where they are only presented to the "general" storage group. Does that sound like what you want to do ?
That's exactly what I tried to describe in my scenario...
One little difference, I will only put the luns (RDMs) needed for the Oracle guests in the storage group for the Oracle cluster... After the migration is finished, I just have to remove them from the "old" ESX storage group. Otherwise, I have to change two storage groups after the migration.
But the idea is the same.
Thanks for your thoughts...