[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[safnog] Prefixes in IGP vs iBGP
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 12/May/15 06:41, Graham Beneke wrote:
>
>
> This raises an interesting question for me:
>
> Why do you choose to distribute this in your IGP rather than as a iBGP
> prefix?
>
> The basic rule is that all customer prefixes should be in iBGP. I have
> typically extended that to be any hosts and services on the network
> are 'customers' of the network (even if they are freeloaders).
>
> IGP is then only used for routing infrastructure and hosts directly
> related to the control plane of the network (route reflectors). On
> smaller networks I've often also made exceptions for operations
> systems like TACACS and monitoring.
We carry customer routes in iBGP. That is still best practice.
I find BGP for Anycast better when you're dealing with having to route
between AS's, e.g., the root name servers.
Within the same AS, BGP carries a lot more overhead to get the same
result, and even though IGP metrics are one of the BGP attributes
considered in the best path selection algorithm, there are several
others steps that could unintentionally come into play.
Internal services that are used by the network (such as NTP and TACACS+)
or by customers (such as DNS) are so basic that I'd like for them not to
depend on BGP, i.e., if BGP falls over, those services should still be
reachable. There is nothing as basic as an IGP; BGP will only work when
you have a working IGP anyway. This design philosophy lends itself well
to imposing IGP-based Anycast. If you perform latency-based IGP routing,
like we do, these services will always be reachable (either by the
network or by customers) from the physically nearest point possible,
because the cornerstone of IGP routing is the metric.
Mark.
-----BEGIN PGP SIGNATURE-----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=f/dW
-----END PGP SIGNATURE-----