[afnog] IP transit and Load Balancing

[Patrick Okui]

Hi Joey, late comment but ...

On  11-Jun-2015 09:17:24 (+0300), Mark Tinka wrote:
> 
> 
> On 11/Jun/15 01:45, Joey ESQUIBAL wrote:
>> 
>> In terms of outbound traffic, can you also recommend what’s the
>> best practice and or ways on how to influence and load balance
>> outbound traffic?
> 
> If you can take full BGP tables, you can use LOCAL_PREF to load
> balance traffic.
> 
> If you can't take full BGP tables, you can ask for partial routing
> from each of your providers + default. But given your border routers
> are geographically dispersed, you'll have limited options on how you
> can load share destinations bound for 0/0 + ::/0.

Even if you can't take full tables you can ask for full tables + default
and filter out prefixes you don't need. Most router oses apply filters
to updates as they come in so they won't wait for the full tables to be
sent. You need to ensure you aren't doing something like the deprecated
"soft updates" though. Crucial to this strategy or Mark's is something
Nishal mentioned.

You must understand where your traffic currently goes currently. Turning
on Netflow on your routers to a collector that can give you a list of
top 10 - 20 destination ASes is very helpful. If you don't already have
that I'd say get it installed first with your network as is.

Last comment is Nishal's approach of controlled deaggregation works best
if you have split your customers evenly across your address space. If
you haven't, then rather than deaggregate all the way to /24s (or /48s)
consider renumbering customers that obtain their addresses via DHCP in a
way that if you split your address space in two you have customers that
can attract traffic in each half.

--
patrick

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 244 bytes
Desc: OpenPGP digital signature
URL: <http://www.afnog.org/pipermail/afnog/attachments/20150615/e58dcf02/attachment.sig>