If I have 15 switches on the same VLAN and each switch has an upper limit on mac-addresses of 16k, and one of them is reaching that upper limit of 16k, what would be the best approach to mitigate this problem? My understanding is that once the switch hits that mark, it will start cycling through removing the oldest mac-addresses first to make room for the new. The problem with this is that drives up CPU, and the switch then spends more time managing its mac-table than forwarding traffic to it intended destination, leading to connection problems for the users connected to that switch. Whats the best way to deal with this aside from getting a new switch that supports more mac's?
-
1You have more than 16K hosts connected to your layer-2 broadcast domain? It sounds like you need to use VLANs and layer-3 separation. A current best practice is to limit a VLAN to a single switch. You can have multiple VLANs on a switch, but each of those VLANs on a switch should not be trunked to other switches. This will isolate STP problems.– Ron Maupin ♦Commented Sep 21, 2016 at 16:49
-
Correct, more than 16K connected at layer2 on a single switch- same BC domain, same VLAN.– user53029Commented Sep 21, 2016 at 17:07
-
That is excessive and unnecessary. How do you connect 16K hosts to only 15 switches? It sounds like you may be trunking VLANs to the switches which are not used on the switches. You can limit the VLANs which are trunked.– Ron Maupin ♦Commented Sep 21, 2016 at 17:10
-
All switchports are patched into WiFi access points, which account for all the hosts we are seeing.– user53029Commented Sep 21, 2016 at 17:22
-
2You probably need to set up separate VLANs with fewer switches per VLAN, and limit the switches to which each VLAN is trunked. As I pointed out above, the current best practice is one switch per VLAN.– Ron Maupin ♦Commented Sep 21, 2016 at 17:26
1 Answer
If you're running into a Layer 2 limit, then instead of tweaking your Layer 2 parameters to handle the unusually large (and probably unsupported) number of hosts, you might want to move up the stack to Layer 3.
If you create IP subnets based on geographic regions to minimise transition from one network to the other, and keep those zones to separate switches, you would in turn decrease your MAC address load.
Theoretically, two subnets will reduce your MAC load by half. Three by a third, and so on.
There could also be an issue with your ARP timers - if your system is keeping MAC addresses around too long, that could be contributing to the problem.