I've been troubleshooting a network for a while where the problem appears to be MAC address table filling up. My customer has two ME3400G-12CS-A switches on two sites connected over a WAN link with local subnets routed between each site. Site 1 acts as a hub and has internet connectivity where Site 2 is a spoke connected to the other site. The issue we're having is that the hub seems to fill up its TCAM tables in peak periods, seriously degrading performance. I haven't been able to figure out why yet as the MAC and unicast tables support 5k entries each (verified in the data sheet and "show sdm prefer") and we should be nowhere near that.
I've been searching through the bug toolkit and googling issues or inconsistencies with the TCAM tables but haven't found any information on the apparent issue, so I believe this is an undocumented defect in SW 12.2(44)SE6. Both switches run the "default" sdm template.
My findings so far:
The hub (Site 1) runs the following:
Switch Ports Model SW Version SW Image
------ ----- ----- ---------- ----------
* 1 16 ME-3400G-12CS-A 12.2(44)SE6 ME340x-METROIPACCESSK9-M
hub#sh mac address-table count | i Total Mac Address Space
Total Mac Address Space Available: 1471
hub#sh mac address-table | i Total Mac Add
Total Mac Addresses for this criterion: 1258
hub#sh platform tcam utilization
CAM Utilization for ASIC# 0 Max Used
Masks/Values Masks/values
Unicast mac addresses: 656/5248 430/3359
IPv4 IGMP groups + multicast routes: 144/1152 10/54
IPv4 unicast directly-connected routes: 656/5248 430/3359
IPv4 unicast indirectly-connected routes: 528/4224 32/175
IPv4 policy based routing aces: 512/512 2/2
IPv4 qos aces: 512/512 37/37
IPv4 security aces: 1024/1024 21/21
The spoke (Site 2) runs the following:
Switch Ports Model SW Version SW Image
------ ----- ----- ---------- ----------
* 1 16 ME-3400G-12CS-A 12.2(53)SE ME340x-METROIPACCESSK9-M
Configuration register is 0xF
spoke#sh mac address-table count | i Total Mac Address Space
Total Mac Address Space Available: 3833
spoke#sh mac address-table | i Total Mac Add
Total Mac Addresses for this criterion: 1145
spoke#sh platform tcam utilization
CAM Utilization for ASIC# 0 Max Used
Masks/Values Masks/values
Unicast mac addresses: 656/5248 157/1175
IPv4 IGMP groups + multicast routes: 152/1216 9/45
IPv4 unicast directly-connected routes: 656/5248 157/1175
IPv4 unicast indirectly-connected routes: 528/4224 25/117
IPv4 policy based routing aces: 512/512 2/2
IPv4 qos aces: 512/512 41/41
IPv4 security aces: 1024/1024 27/27
Summary:
hub: 1471+1258 = 2729 (almost 3359 from TCAM utilization)
spoke: 3833+1145 = 4978 (almost 5k/5248 from TCAM utilization)
The hub in total only seems to support 3k MAC addresses whereas it should support 5k. The only differences between the switches I can find are different HW revisions and SW versions. Is this most likely a software bug in 12.2(44)SE6? Is it only cosmetic or is it the cause of the degraded capacity?
Any comments?
