I have a five node hyper-converged cluster running. For switches I have two Juniper EX4600 and two EX4100 switches in two virtual chassises. Each node has 2 x 1G to each EX4100 in LACP for corosync and 2 x 10G connections to each EX4600 for VM and Ceph traffic.
I'll be replacing the cluster soon and I want to know if there's a better way to configure networking. My problem is that virtual chassis hasn't really been all I thought it was. Juniper says you can use NSSU to upgrade each switch in a stack without losing service, but as I've found out, it's only allowed in *very* specific configurations, and even then most people say it never works.
As it is, I have no idea what would happen to the cluster if I upgraded the switch and restarted it, taking around 10 minutes for the switches to reboot...
Does anyone run their cluster with redundant switches and can give tips on a good configuration?
I'll be replacing the cluster soon and I want to know if there's a better way to configure networking. My problem is that virtual chassis hasn't really been all I thought it was. Juniper says you can use NSSU to upgrade each switch in a stack without losing service, but as I've found out, it's only allowed in *very* specific configurations, and even then most people say it never works.
As it is, I have no idea what would happen to the cluster if I upgraded the switch and restarted it, taking around 10 minutes for the switches to reboot...
Does anyone run their cluster with redundant switches and can give tips on a good configuration?