Unfortunately my ea6350v3 continued to crash/reboot like clockwork as described here.
After installing the corresponding mainline OpenWRT 21.02.1 r16325-88151b8303, all other things being the same, the problem goes away; it's been up for three days and used as my primary access point and a managed switch without a problem.
Unfortunately, by giving up the NoTengoBattery build I lose more than 100mbps of wireless speed on the 5GHz band; again, all other things configured the same. Edit: correction, NOT the same. Installing irqbalance closes the gap. Reducing cpufreq up_threshold
from 95 to 65 to compensate for the reduced sirq thanks to balancing the interrupt affinities gets it much closer to your build's performance. It also means I can't use luci to manage the switch, but that's less of a problem; the config switch_*
settings in /etc/config/network
get the job done.
One interesting issue (which I don't think is a problem on the NoTengoBattery build, but I'm not sure, I'd have to boot back into it and double-check):
While I can get the "WAN" interface ("port 5") to act as part of the main switch easily enough, and it works in VLAN groups with the other four ports for most ordinary TCP/IP purposes, it doesn't seem to work with Chromecast, which relies on mDNS. On other words, with this setup
VLAN 1: 0t 1 2 3 4
VLAN 20: 0t 1t 5
A chromecast device attached to port 5 cannot be found on VLAN 20 (port 1 is the trunk port in this case). It otherwise has functional access to the network and can be pinged from other devices; it's just not visible to those devices as a chromecast target.
If I move one of the "LAN" ports into VLAN 20, e.g. port 4:
VLAN 1: 0t 1 2 3
VLAN 20: 0t 1t 4 5
And move the Chromecast device to port 4, the device immediately becomes visible as a cast target to everyone on VLAN 20. I can only speculate as to why this is, I suspect it's an L2/L3 issue.