Hello,
first of all, I cannot open a JTAC case to report this issue.
I hope the following problem report still makes it to the support department, so this can be fixed.
I noticed on EX3300 (12.3R12.4) and at least EX4550 (13.2X51-D35.3) platforms as well, that in specific situations some MAC addresses are either forgotten, never learned or the hosts are suffering from a high latency.
Some days ago I experienced some strange issue in a layer 2 network, when one specific host was not pingable in one rack, but if the host was started in another rack everything worked fine. The VLAN and port configuration was identical.
When debugging this problem, I noticed that at the ToR switch where the host did not work, the command show ethernet-switching mac-learning-log showed a "learn" of the MAC address followed by an immediate "delete".
Okay, so probably the MAC table was saturated for this specific MAC hash? and the MAC address could not be learned. I increased mac-lookup-length from default to 8 and the host immediately sent replies to my pings. Great. If the pings would have been low, I would have been happy, but actually I had about 30% packet loss and the latency was at about 180 ms.
Finally the problem was solved by removing the configuration part:
ethernet-switching-options secure-access-port mac-move-limit 2 action log
I had issues with ethernet-switching-options secure-access-port mac-move-limit X action log some while ago, so I gave this a shot. To me it seems like mac-move-limit produces issues for some MAC addresses at least when used together with an increased mac-lookup-length. Clearing the affected MAC address on the switch did not bring any help by the way.
Kind Regards,
Leon Kramer