[PDB-tech] proposed new attribute: max_allowed_peering_next_hop_latency

[Job Snijders]

Hi Eric,

On Wed, Apr 20, 2016 at 07:31:47PM +0000, LOOS Eric (BCS/CBU) wrote:
> I am not 100% familiar with the internals so just to make sure I
> understand; this attribute is intended to signal to prospective peers
> (as part of a peering policy) that you have as a network a particular
> max latency peering policy. As far as I understand it, you would not
> request a network to document the latency every particular peering
> record has to the actual next L3 hop, correct?

Correct. Its just the value you'd see by typing 'ping $public_peer_ip'
on your peering router.

> I understand the comment about lawyering over a definition of remote
> peering, but I don't think it is that much of an issue. Remote peering
> is very much an accepted way of connecting to IX'es and it has many
> valid use cases. Furthermore 'are you using remote peering' is a very
> understandable question for new participants (which are most likely
> not to meet the requirement). Therefore I would indeed stick with an
> easy definition of a policy requirement saying remote peering:
> 'accepted, not accepted, no preference'

But that removes the visibility into what the actual latency is and
might muddy a decision. If my record states "10 ms" and you ping my
public peering IP on say AMS-IX, and you notice that the latency is more
then 10 miliseconds, that tells you something useful without engaging in
what defines remote peering or not.

> In addition, again if I understand it correctly, this only works one
> way? i.e. someone checking whether they can peer with NTT will find
> the max_allowed_peering_next_hop_latency attribute and have to decide
> whether they can connect, it doesn't work from the point of view of a
> network seeking to peer with all non-remote peering members on an
> exchange? That would also be very valuable but calls for a remote
> peering record on each IX connection. Luckily, this could be
> maintained by the IX'es in most cases because they know whom came in
> via a partner.

I think this should be discussed in a separate thread, as its a separate
alteration to the data model.

> As such, indeed it is a policy requirement, but it is important enough
> next to ratio requirements, multiple locations, ... given the amount
> of new customers joining in this way.
> 
> Just my 2 cents :)

Thanks for your feedback.

Kind regards,

Job