I'm no networking expert which is probably why I'm really confused here...
An ESX host with 2 NICs was running fine with 2 Etherchanneled NICs and "route with IP hash".
We changed the routing policy from "ip hash" to "virtual port ID" and then tested a few machines.
It's not clear if we tested the wrong machines, or if the problem did not manifest for 24 hours, but after a full day of no reported issues, the morning of the second day, many suddenly had problems reaching some of these VM's.
For any given client system, it was hit and miss (Cisco switch path A success, switch path B fail, etc.). Some locations couldn't hit anything.
We then changed the routing policy back to "IP hash" and everything was happy again.
I'm a "network noobie" but I'm struggling to understand where the breakdown was here. I understand that the "traffic segmentation" changes from a pre vNIC basis to an IP conversation basis between these two modes. But I fail to see why this would change behavior for a given network route. I wouldn't expect EtherChannel itslef to be the issue here (traffic is still assigned to different pNICs but just based on different rules) but I'm not discounting anything at this point.
I'd be very interested as to why switching to the default "route by virtual port ID" could cause network failures that could be corrected by "IP hash".
Thanks!
