[afnog] IP transit and Load Balancing

[Mark Tinka]

On 16/Jun/15 09:05, Andy Davidson wrote:
> My advice is to try to understand, using Netflow or similar, who your
> largest inflows of traffic come from. In my experience, heavy inbound
> networks have well over half of their traffic from under ten
> peer-ASNs. Once these are understood and mitigated, it’s much simpler
> to deal with traffic engineering, as balancing for those larger
> senders will essentially wrap up the majority of your traffic. My
> understanding of this reality is quite European slanted but I’d be
> really happy to take a look and send some advice about your top
> traffic originators either on list or off-list - and wearing no hats.

I suppose the biggest issue the OP will have with this is the fixed
nature of his interconnect to each upstream, i.e., the STM-1's and STM-4's.

I'm not sure about how much traffic the OP's network is
ingressing/egressing from/to his upstreams, but at some point, traffic
engineering becomes restricted by the physical limitations of the
circuit. Once he gets to that point, the only option is to increase that
interconnect bandwidth. Considering the OP sounds like he's purchasing
long circuits from Benin to Europe, this might not be always easy.

Mark.