[afnog] Internet Traffic distribution management

Frank Kuse rombokite at hotmail.com
Wed Sep 30 09:36:28 UTC 2015


Hello Mark,

Yes we do have discrete eBGP session with each peers and your proposed option of Lagging could help in a long way towards minimising the number of peering eBGP sessions. Meanwhile I will share the recommendation to the team as we look forward to changing our core routers and having bigger uplink pipes.
Once again thank you for the contribution.
Regards,

Frank

Sent from my iPhone

> On 30 Sep 2015, at 06:06, Mark Tinka <mark.tinka at seacom.mu> wrote:
> 
> 
> 
> On 29/Sep/15 16:59, Frank Kuse wrote:
> 
>> 
>> 
>> Each link is a minimum of STM16.
> 
> I'm assuming that you have a discrete eBGP session per STM-16 link.
> 
> If so, why don't you reduce the number eBGP sessions so that instead of 8x, you have 4x? You can do this in one of two ways:
> 
>     a) Run a POS LAG with your provides to whom you have multiple links. Assuming both your router and theirs can support this, you end up with a single logical 
>         link that is equal to the sum of each individual STM-16 combined. This way, routing becomes easier and you end up with a larger link that will support ECMP.
> 
>     b) The other option is to run eBGP Multi-Hop, running the BGP session across a Loopback interface on either side, and using static routing on both sides so that 
>         you load balance traffic across all links to/from the same upstream. Again, this has the benefit of having a single eBGP session for all links to the same 
>         upstream, while providing ECMP.
> 
> With the increased capacity and the halving of all your eBGP sessions, you can have an easier time doing traffic engineering using classic methods such as AS_PATH prepending, safely de-aggregating and/or discriminate announcing of prefixes amongst the providers.
> 
> Hope this helps.
> 
> Mark.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.afnog.org/pipermail/afnog/attachments/20150930/f387a336/attachment.html>


More information about the afnog mailing list