[ixpmanager] Techniques to influence inbound (download) traffic across multiple IXPs with /24 prefixes

[Salvador Bertenbreiter]

Very helpful! Thank you for your advice Douglas :)


El El jue, 8 ene. 2026 a la(s) 06:57, Douglas Fischer <
fischerdouglas at gmail.com> escribió:

> If there were two networks that met not through IXPs, but through their
> own links... How would this be resolved?
>
> MED - Multi Exit Discriminator
>
> For this to work over IXPs, 3 requirements need to be met.
>
> A) Transparent MED. Route-Servers need to forward the received MED.
>
> B) From the point of view of the participants sending routes to the
> Route-Servers:
> B.1) Stop the WRONG methodology of forcing more specific routes to be
> advertised to Route-Servers, as well as excessive prepends.
> B.2) Apply MEDs that reflect the preference for each path.
> Note: I really like the scenario where the MED is used as Cumulative
> IGP-Cost, and the IGP cost reflects distance in kilometers, or latency in
> microseconds.
>
> C) From the perspective of participants receiving routes from
> Route-Servers:
> B.1) Stop the WRONG methodology of indiscriminately increasing local-pref
> for all routes received from Route-Servers.
> B.2) Start comparing and analyzing the MED for eBGP, especially in IXPs,
> activating parameters such as "always-compare-med",
> "compare-different-as-med", and "deterministic-med".
>
> Part "A", from the route-servers... is easy!
> Parts "B" and "C", which depend on network operators ceasing to look at
> their own navels, are more difficult.
> I only end up seeing this level of maturity in large carriers.
>
> Em qua., 7 de jan. de 2026 às 13:44, Salvador Bertenbreiter via ixpmanager
> <ixpmanager at inex.ie> escreveu:
>
>> Hi everyone,
>>
>> Hope you’re doing well. I’d like to ask how you approach inbound traffic
>> engineering (download traffic) when you’re connected to multiple IXPs (some
>> local, some remote) and you only have small prefixes (e.g., /24), so
>> announcing more/less specifics isn’t really an option.
>>
>> In that scenario, what methods have worked for you besides simple AS-path
>> prepending (e.g., no prepend on local IX, prepend 1 on the closest remote
>> IXP, prepend 2 on the next, etc.)?
>>
>> Thanks in advance for any guidance or real-world examples.
>>
>> --
>> Salvador
>>
> _______________________________________________
>> INEX IXP Manager mailing list
>> ixpmanager at inex.ie
>> Unsubscribe or change options here:
>> https://www.inex.ie/mailman/listinfo/ixpmanager
>
>
>>
>
> --
> Douglas Fernando Fischer
> Engº de Controle e Automação
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.inex.ie/pipermail/ixpmanager/attachments/20260108/c38ed7d1/attachment.htm>