ARP cache needs clearing constantly

Marty Harrison · ‎02-21-2014

I help manage a WISP network which has a Cisco 3750 at its core and a couple of Cisco 7200 routers to two separate ISPs. We have configured public and private VLANs. The infrastructure is based on point to multipoint links with cisco SG300 switches at each main location. I have a problem where devices are not contactable until the ARP cache on the router is cleared or they are pinged from a device on the local LAN. It seems that as soon as the switch times out the address from the CAM tables, it will not respond until the ARP cache is cleared. A simple example is this.

router 10.201.1.1 --- Core switch --- switch 10.201.2.254 --- AP 10.201.2.55 --- CP 10.201.2.37 --- Customer x.x.102.37

All devices remain active up to the AP. The CP and customer will drop off the network randomly. I would expect the traffic to refresh the CAM table as per normal. The ARP cache on the router shows the exact same information before and after clearing so there is no spoofing or eroneous info there. This is not an isolated case, it is exactly the same across all of the other devices in the setup.

This has only become apparant since moving the devices to being managed in VLAN 201. prior to that, the client had everything (over 3000) devices in VLAN 1 with public IPs along side Private. Obviously a problem.

Anyone seen this sort of problem and able to offer any tips, ideas or suggestions? I am all out of ideas.

Thanks.

Marty

Marty Harrison · ‎02-22-2014

Just to add to this. I have read a lot of information about HSRP and having to synce CAM and ARP tables. I am not using HSRP and in this case, the second 7200 router is connected to a different ISP and operates completely separately.

In light of the information on HSRP. I have reduced the ARP timer for the two VLANs to 240 seconds, which being less than the CAM timer of 300 seconds means that the CAM table always remains up to date. I see this as a work around because it will obviously lead to a large amount of ARP traffic but it does seem to fix this problem. I am not sure why when the CAM table loses the MAC address, fresh traffic doesn't reach the switch and trigger the update of the CAM table. It seems like the traffic is not being sent to the right place.