[nanog] Connections among ASes (fwd)

Fri Nov 30 04:28:16 UTC 2007

On Thu, 29 Nov 2007 20:49:13 CST, Chengchen Hu said:

> Suppose the following example. ISP A has a router A1 in IXP1 and a router A2 in
> IXP2; and ISP B has a routers B1 in IXP1 and a router B2 in IXP2. It is
> possible that we have DIRECT link A1A2 and B1B2 to connnect two IXPs, but I
> don't think there may be DIRECT link like A1B2 or A2B1. Since it should be much
> cheaper and easier for ISP A and ISP B to be connnected in the same IXP using
> links like A1B1 or A2B2. Am I right?

You're quite possibly wrong - for instance, if I'm ISB B, I *might* want
to have a direct peering session between A1 and B1 - but *also* have a
connection from B1 over to the *other* router A2, for several reasons:

1) I may know that the *next* router hop after A1 has questionable reliability,
and thus I want a fall-over to A2, which has better connectivity upstream.

2) I may be able to get a second link over to A2 for "essentially free" because
I have a connection to IXP2 because I peer with *another* provider C (who I
have to connect at IXP2 because C has no presence in IXP1).  At that point,
I may be able to get B1-C2 for some cost - and then B1-A2 as a backup to the
B1-A1 is almost free at some IXPs - just one interconnect across the room.

3) Due to traffic balance quirks (maybe I'm content-heavy at IXP1 and
eyeball-heavy at IXP2), I may qualify for peering at one IXP but not the
other - so if my only peering is at IXP2, I have to haul traffic from
IXP1 to IXP2 and peer there. (Yes, that *would* be odd - but I've seen
stranger stuff happen with peering.. ;)

4) Traffic engineering may indicate that doing a cross-connect may be
faster/better - if you have a lot of traffic from your AS hitting router
A1, but the *other* end is just upstream of B2, you have 2 choices:

a) dump the traffic from A1 to B1 and let them haul it to B2 (hot potato
routing).  This can suck if B1-B2 is congested...

b) Put in your own link from A1 to A2/B2 - this can win if your A1-B2 is
less loaded than B1-B2 is.

I'm sure that the guys who do the traffic engineering thing for a living
can come up with even more examples why it may be different...

> So in your case, all the suppliers and peers DIRECTLY connected to any one of
> your routers are located in the same IXP?  

In our case,  we have routers inside our AS1312 - and our two main next-hops
(Level3 and NetworkVirginia) are both at the other end of many miles of fiber.  
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 226 bytes
Desc: not available
URL: <http://mailman.nanog.org/pipermail/nanog/attachments/20071129/9c55f853/attachment.sig>