** I really appreciate you taking the time to read and answer my question but please read this first -> I am not running SCVMM - I DO NOT WANT TO AND CAN NOT RUN SCVMM - Please bear this in mind **
Hi guys,
I am currently exploring Hyper-V clustering and have some questions.
This is what I have already:
- Lots of cluster validated hardware.
- A totally working and bullet proof iSCSI storage network.
- A totally working and bullet proof iSCSI SAN running over 10GBE.
- A fair bit of systems and network experience, but not with the Windows.
For testing I have:
2 nodes in a working cluster with iSCSI quorum.
Node A: Has 3x CSV, each CSV is running a single VM.
Node B: Part of the cluster, but not running any disk or VM resources.
Perhaps because I don’t fully understand the CSV architecture, I have only 'connected'\mounted my ISCSI volumes on node A.
* My network setup looks like this:
- 2x LACP 10GE, bonded interfaces just doing iSCSI. I cannot do multipath due to the way VLT is deployed on my storage switches.
- 2x bonded 1GE, again LACP, but this time 2x stacked switches, one 'sub interface' for management and domain traffic, one 'sub interface' used for a hyper-v virtual switch\VM traffic, same physical interface.
Windows 2012 R2 across the board.
This is what I *want* to be able to do:
- Have an Hyper-V cluster of X nodes, which I can grow in-line with my requirements. i.e. up to 10 nodes, or whatever.
- Always keep enough spare capacity, in terms of CPU and RAM (storage is obviously abstracted), in the cluster to deal with a single node failure.
- I would ideally like this capacity to be dotted about the cluster, not contiguous capacity on a single node. It does not have to be this way, but it would be great if it could...
- If a node fails, have the cluster bring up my CSVs and clustered VM roles around this cluster, where there is CPU and RAM capacity.
- I appreciate SVMM is likely to care of this resource based placement, and the CPU thing is not essential - just deciding on placement based on there being enough RAM to start the VM would be great.
My cluster is working, but I am a little confused about how CSV works. If could give two scenarios I am unclear on (with my network\lab set-up as above):
Scenario One.
- If I 'move' a VM from node-A to node-B, what is actually happening here in terms of the VMs storage?
If the CSV is still owned by node-A, but the VM is now 'owned' by node-B, does disk I/O now traverse my 1GB network? i.e. down to node-A (the CSV owner node) *then* down to the iSCSI LAN, with return iSCSI traffic also doing the same? i.e. is the CSV owner node acting as a proxy for I/O?
I appreciate I can designate networks for CSV traffic - but even if I were to do this, forcing CSV traffic through my 10GB iSCSI network doesn’t feel right, presumably it still has to hit the CSV owner node and I don’t fully understand what impact this would have on latency etc.
Scenario Two.
Same as above, but If I connected (in terms of iSCSI target) every single CSV I ever provisioned on every host, from now until the end of time - how would scenario above play out? The CSV on which my VHDX resides is still owned by node-A, but node-B (technically) has a path to the iSCSI LUN over my 10GB network. The LUN is not mounted, it doesn’t have a mount point, it is CSV. What would happen here?
Some further questions:
- If I create a CSV on one node, then a second CSV on another node, am I obliged to connected the iSCSI targets on all hosts?
- Is it possible to live migrate a VM between two CSVs, so I can ensure disk I/O traverse the correct network? Is there a better way of achieving what I want?
- Without SCVMM, what dictates where resources end up in the event of a cluster node failure? Take this scenario, for example:
Node A:
Total RAM 10G
RAM being used by parent partition: 2GB
RAM being used by VMs: 3GB
Available RAM: 5GB
All VMs on single CSV owned by this node.
Node B:
Total RAM 10G
RAM being used by parent partition: 2GB
RAM being used by VMs: 3GB
Available RAM: 5GB
All VMs on single CSV owned by this node.
Node C:
Total RAM 10G
RAM being used by parent partition: 2GB
RAM being used by VMs: 3GB
Available RAM: 5GB
All VMs on single CSV owned by this node.
... let's say between all those nodes there are 9x VMs using 1GB of RAM each, one of these nodes explodes, how are the clustered resources distributed amongst the remaining nodes? Can I control this? How can I understand the logic behind this?
Basically the two features I want are:
- Automatic recovery of VM resources in the event of node failure.
- Live migration.
But I need flexibility in terms of VM deployment, at the point of creation (as you may have guessed)
Bit of a long post, but any thoughts or help would be great.
Cheers.