IPv6 BGP TE (was Couldflare routing problems)
Chris Welti
chris.welti at switch.ch
Thu Jun 21 18:23:38 CEST 2012
Hi JF,
Am 6/21/12 4:24 PM, schrieb Jean-Francois.TremblayING at videotron.com:
> Hi Chris.
>
>> I specifically said I don't see a valid use case for *globally*
> advertising
>> more specifics. Locally is a different case:
>> Using more specifics for TE is fine towards your upstreams or direct
>> peers, where you can still negotiate acceptance of more specific
>> announcements for this purpose.
>
> Glad to see your local policies could be different. This will at least
> allow
> for your clients to load-balance.
>
> What about a client doing load-balacing between multiple providers? Would
> you,
> for example, carry longer prefixes from your clients one or multiple peers
> away?
Only if it makes sense to do so. E.g. for a paying customer that is multi-homing
within our AS (as in redundantly connected to us) we will of course carry the
more specifics internally, *within* our own AS. I surely would not expect any
other upstreams or peers to accept longer prefixes. Therefore, there should
always be a covering /32 announcement with a valid path to the longer prefix.
There can be special arangements with direct peers to accept longer prefixes
in case of a mutual multi-homing customer, but I wouldn't consider that the norm.
> It boils down to what do you consider local vs global. You'd probably have
> to
> put a limit on the AS path length of the longer prefixes.
>
> I guess this all can be negotiated with transit and peers, but it's good
> to share views on this so that we end up with consistent policies
> wherever that's possible.
Well, the consistent global policy is that prefixes up to the minimum-allocation size
of the RIRs are accepted.
Regards,
Chris
More information about the ipv6-ops
mailing list