Yup! Let's do it. Good Policy.
From: MICE Discuss <[log in to unmask]> On Behalf Of Richard Laager
Sent: Monday, October 28, 2019 3:25 PM
To: [log in to unmask]
Subject: [MICE-DISCUSS] MICE MAC Address Limit
We have discussed this a bit in the past, and this came up at the last UG. I'm looking for feedback from the group before formally proposing this to the board for a vote.
I first brought this up to the technical committee, CCing the board. I have heard no objections. Jeremy is "in favor of it 100%".
Based on our quick conversation after the UG, I _think_ Anthony is in favor as well; we both made notes to follow up on this.
I am proposing that MICE change its MAC address limit from 5 MACs per port to 1 MAC per port. 1 MAC address is enough for normal scenarios and this would further limit the potential for problems on the fabric. This restriction is one that SIX and AMS-IX both have, with the latter being known for their excellent configuration guide for participants:
To be clear, this is still only for end ports, not ports facing remote switches, of course. The remote switches are responsible for enforcing the current MAC limit, and would be responsible for changing those settings as described here.
If someone is swapping their MICE-facing router, the resulting port flap on the MICE/remote switch will clear the limit anyway, if they are directly connected. If they are unable to flap the port (e.g. because they have someone else's layer 2 gear in the middle), they will have to either work with that carrier or their switch operator (MICE/remote) to flap the port. This is a non-zero burden, but I think it's pretty minimal in practice. Additionally, if someone plans to make an equipment swap, we would increase the limit in advance, temporarily, upon request.
For reference, SIX goes further and locks you to a particular MAC address in an ACL, so you have to contact them to change your MAC. I am not proposing that. Still, I've been through that a couple of times, and even that isn't really too big of a deal. So I don't think the change to a limit of 1 MAC at MICE will be particularly onerous.
There are a couple of participants who have two IP address sets (one
IPv4 and one IPv6) on the same port. For those ports, the limit obviously must be at least 2 MACs, but I propose 3 MACs as the limit. If they're using two IPs, they probably are doing so to get some redundancy out of it. For this particular use case, needing to flap the port defeats that redundancy goal, as it breaks the other router. Having a MAC limit of 3 would allow them to swap a device without needing to flap the port. We would grandfather these setups until they are looking for a port-type upgrade; at that point they would need to get to one IP address set per port (by splitting into two ports or reducing to one IP address set) or request an exception as outlined below. The exception process would allow us to better evaluate this use case at that time.
There are some members who have multiple MACs showing up now, despite using only one device. For those, I am proposing we set the MAC limit to however many MACs are currently showing up on their port. That is, they would be grandfathered at their current situation for now. At the time of this change, we would encourage them to investigate and fix the multiple MACs issue at their leisure. In the future, if they are doing a equipment swap (especially like-for-like), we would again encourage them to address it, but would otherwise leave their higher limit in place. If they do a port-type upgrade, we would not bring the grandfathering over.
If they needed, they could ask for an official exception at that time.
There may be cases where people cannot reasonably fix their routers to use only 1 MAC, other scenarios I haven't considered, or things that may come up in the future. For that reason, I think we should allow for the possibility of exceptions upon request to the board. In practice, I would expect the board would confer with the technical committee and/or the technical committee would be bringing these to the board anyway. The goal is to strike a reasonable balance between protecting the fabric without preventing networks from connecting.