Hyper-converged Cluster Networking

[smyers119]

What is the best practice for the corosync network? I can't imagine it having much bandwidth requirements, and I'd hate to use a seperate nic just for this. seems to just need low latency.

Since I have ceph public and cluster networks separated on separate 10G Nics, Can I put corosync on the ceph cluster network? It has sub 0.1millisecond latency between the servers.


Thanks

 

[sb-jw]

Since I have ceph public and cluster networks separated on separate 10G Nics, Can I put corosync on the ceph cluster network? It has sub 0.1millisecond latency between the servers.

For years I have only been running HA clusters with 2x 10 GbE connections and even without a second ring for Corosync. So far I've never had any problems with it. However, you have to be aware that the utilization of the links may also lead to the failure of the cluster because the latency increases.

If you really want to do it right, then have separate ports for management, migration, storage, cluster, internal and external. You should also plan your own switch pairs for storage and Corosync. This means that a node would then have 12 links and you would have at least 6 switches (two each in a network) in the rack.

From my point of view, it makes more sense to ensure that the 2x X GbE links per node are higher than the external bandwidth (so as not to lose storage and management etc. in the event of a DDoS attack) to the Internet. This measure alone could be enough. Otherwise, a single 100 Mbit switch for the second ring could possibly be sufficient for Corosync.

Ultimately, it depends on your sense of security for the setup or the availability requirements or budget.