[afnog] local pref issue with HE

[Mark Tinka]

On 25/Dec/15 18:18, Frank Habicht wrote:

> Hi Mark,
>
> I'm seeing a good potential for suboptimal routing.
>
> If you have a Provider TransitA connected to your East and a provider
> TransitB connected to your West, and interconnection points far apart in
> geography and latency, and TransitA also covers the location in your
> West, then any traffic from on-net services and single-homed customers
> of TransitA in the West will surely be routed to the East first before
> given to you.
>
> Which should cause some unnecessary latency, likely avoided by getting
> service from TransitA in the West as well.
>
> There might be plenty of other valid reasons to choose that setup
> [ even interfaces cost $$$ ], but in my understanding that's a negative
> aspect that remains.
> And one should be aware of it.
>
> If there's a general way to work around it, I'd be interested ;-)

If you mix-and-match providers of different sizes together, you are
bound to run into this problem.

If, however, you have multiple providers of the same size at the top
tier of the global scale interconnected to you at different locations
within a reasonably similar region (take Europe, for example), more
often than not, their customers (other ISP's) that connect to them are
multi-homed to at least two or more of the same providers you also
connect to.

That is how you achieve the relative symmetry.

Any latency problems can be easily resolved with LOCAL_PREF adjustments
at the interconnect point with the provider that gets you there best,
assuming peering does not do that already.

The situation becomes significantly easier to control if you peer at
more than half of all the major global exchange points, offloading
two-thirds or more of your traffic there. In this case, the actual value
added by your transit providers is - albeit important - marginal.

Mark.