Re: draft-amante-isis-reverse-metric-01

On Dec 7, 2010, at 09:43 , <bruno.decraene <at> orange-ftgroup.com> wrote:

> [Warning: thinking out loud...]
> 
> a) I understand the need to cost out a link with a single CLI on a
> single router.
> b) I hear that some would prefer management action be done with
> management protocols rather than routing protocols extensions.
> 
> So what about doing (a) with management protocols? i.e. a router A
> providing a single CLI and triggering the remote action on router B
> using management protocols.

That sounds much more complicated than what the draft suggests, and doesn't even solve everything what the
draft solves ('last-resorting single node on a multi-access LAN'). First it requires manager-side (As
opposed to the agent) of a management protocol (SNMP, NETCONF, you-name-it). Second, either manager- or
agent-side would have to perform quite sophisticated work to find out appropriate interface that needs
to be altered (since we don't error-prone humans to do it). And finally, there must be then place where
operator could see why on the earth metric of an interface isn't the same as in the config (easy with
'Reverse Metric TLV' - it's in the neighbor info).

Also looking from completely different angle, there is one application for Reverse Metric TLV that cannot
and should not be done via manual or automated config management. I've already mention it when topic first
popped up several weeks ago - Reverse Metric TLV can nicely solve problem of LDP-IGP sync on multi-access
LAN. For P2P now routers just set overload bit or max-out intf metric, but that doesn't work on
multi-access LAN because existing solutions are too coarse for it. Reverse Metric TLV provides fine
granularity that allows quite elegant implementation of LDP-IGP sync on multi-access networks.

Kind regards,
iLya