Wednesday, March 28, 2018

Changing out core L3 switch stack

Hoping someone can sanity check me here - I'm not a network guy by trade, but it's one of my hats at this job. We have an older Juniper EX4500 stack that is being decommissioned and replaced with QFX5100s. I've already moved over all the physical connections and have a trunk between the switches for all the VLANs.

The only thing left to do (and it's a big thing) is to get the routing interfaces (IRBs/RVIs in Juniper parlance) moved over to the new stack. We have 33 VLANs with gateways provided by the Juniper (172.16.1.1, 10.10.1.1, etc. etc.). All the actual phyiscal connections are on the new switches, but the network interfaces for those IP addresses are on the old switch stack, so any routed communication has to traverse the trunk back and forth. There just a few manual static routes, no dynamic routing.

My plan was to add the IRBs and routes to the new switch and then commit the changes at the same time that I unplug the old switch. I tried that early this morning, gave it about 6 minutes, and had to rollback when nothing was coming up. I was connected to one of the VLANS on a Win10 box and couldn't ping it's default router anymore. We are a 24/7 business with critical public safety systems, so I couldn't sit and tinker around with it.

Assuming I didn't muck up the config and everything was correct, what's your next best guess for why this didn't work out? Was it a terrible plan from the start? ARP refresh timer is at the default 300s, but I assumed when the new switches got the routing interfaces, they would broadcast a gratuitous ARP and everything would pretty quickly update.



No comments:

Post a Comment