Hi.
I'm running a Hyper-V 2019 2-node-cluster using the VSAN free Windows binary directly on the host and I'm not really happy with the network performance. This is due to Microsoft really messed up the implementation of the network stack in 2019. There are a couple of very good explanations on this on Spiceworks
https://community.spiceworks.com/t/serv ... nce/724968
and I can totally agree to that. The worst part is the vNIC in the ManagementOS, which is sort of a light version of a real vNIC with only limited capabilities and throughput. When I built the cluster I wasn't aware of this, thus I used ManagementOS-vNICs a lot, also for VSAN, because I only have limited NICs available.
Here is the current setup (both nodes are identical):
1x 10G pNIC for VSAN sync
1x 10G pNIC in conventional vSwitch for VPN
2x 10G pNIC in IOVenabled SET-vSwitch
All pNICs are Intel X553.
The conventional vSwitch is needed, because SRIOV enabled SET switch doesn't allow MAC address spoofing, which is required for VPN (I tested it).
At the SET switch, there are several ManagementOS-vNICs connected
2x VSAN sync vNICs
1x Heartbeat vNIC
1x LiveMigration vNIC
1x Management vNIC
Even with all the tuning tips from Spiceworks, I barely get more than 1G on these vNICs which is painful, because there is a file server running on the cluster. So, the idea is to move VSAN to a VM to use better vNICS and also free the VSAN sync pNIC for management and file server.
What are the recommended steps to do this? I would
1. pause one cluster node
2. Stop and deinstall VSAN.
3. install CVM
4. configure CVM to have the exact same iSCSI-targets
5. resynchronize with other cluster node
6. repeat for second cluster node
Here are my questions:
1. Is this working or am I missing something? This is a production system and I don't want to lose data.
2. What is the correct ALUA setting? Both nodes are identical, so ALUA should be off, right?
Thank you for any hints.
Martin
The Latest Gartner® Magic Quadrant™Hyperconverged Infrastructure Software