MikroTik

Over the summer we upgraded the majority of our tower routers to CCR16s. In the last 60 days we stared receiving calls from some customers stating their connection suddenly stopped working. After checking the link from the tower to the customer we discovered the problem was with the ARP entries in the CCRs. When we compared the DHCP leases to the ARP entries we noticed that a different customer was ARPing to the address that was assigned to the customer who called. The only way we have found to resolve this issue is to manually remove the entry belonging to the wrong customer. Once the duplicate entry is cleared the customer who called can connect. (I have attached screen shots of one of the occurrences.)

We are seeing this problem on every version 6 revision we have tried; even 6.6. When we did our upgrades there was no configuration changes between the previous MikroTik routers and the CCRs. The only difference is the ROS version that is running (and of course the hardware). The previous routers that were in place where RB1200s running v5 code.

We did see that in ROS 6.5 an option was added to disable conflict detection in the DHCP server. I have looked to see what this provided, but I have not seen any information other than the brief announcement in the change log. Was this added because of this problem, and if not what is is for? How is this enabled?

This problem is generating several calls a day and is putting stress on my support crew, who in-turn is looking for a resolution and answers.

We are using Canopy and Ubiquiti APs and CPEs.
[edit] Submitted as bug [/edit]

Looks like a classical concurrency programming error to me, and not a DHCP server error. I guess the ARP code isn't synchronized correctly.

If this is an issue with ARP code not synchronizing, then I gather it is a bug with v6 and should I open a support ticket? In the mean time do I need to look elsewhere for a work around/solution?

Yes, you should open a ticket for this in any case. MT might be able to suggest a workaround, I don't have a good idea, you can't pin a NIC IRQ to a CPU on the CCR/Tilear architecture, can you?

Thanks for the info. I have submitted this issue.

you can't pin a NIC IRQ to a CPU on the CCR/Tilear architecture, can you?

Unfortunately I can't.

Stall ARP entry issues are still plaguing Mikrotik ROS 6.X 4 years later...