Re: Question on the topology of Internet Exchange Points

[Paul Vixie <paul () vix com>]

> > in other words there appeared to be no "exchange-based topology", more
> > like a "hybrid exchange and PNI topology."
> >
> > Paul Vixie
>
> It is interesting. Is this the common case for the IXP infrastructure?[1] I
> mean the hybrid topology? It seems that it is both directly-connected and
> exchange-based, but the direct links between participant ASes are much more
> the links to the central switch. Right?[2]  One more question is: if one AS
> participate in an IXP, is it necessary for the AS to place a router in the
> IXP?[3] In other words, all the participant ASes must have at least one
> router in the IXPs (these routers are geographically nearby)?[4] Can they
> use remote connection?[5]
>
> Kai

[1] i think it is very common for an IXP to have a switch, to have racks for
participating network operators, and to allow PNI's between those participants.
however, i know of IXP's which only have a switch and require the participants
to connect to it via telco circuits; and IXP's which have racks but which do
not allow PNI's between participants.

[2] at PAIX when i was there, the PNI's outnumbered the switch connections by
more than 100:1.  i don't know what's true today, or elsewhere, but in the
IXP's where ISC has a presence, we generally have between two and twenty PNIs
and only one or two switch connections.  i'm assuming that this is common.

[3][4][5] at PAIX when i was there, all switch and PNI connections had to be
to locally installed routers.  we didn't allow crossconnects between telco
equipment (because we were an IXP not an MMR), and we didn't allow connections
from a telco equipment to our switch (because we were an IXP not a MEP).  the
"network effect" at an IXP only comes when everybody has skin in the game, a
real BGP-speaking router of some kind.  i think PAIX (now part of S&D) still
has the same rules.  i don't know what's true of other IXP's around the world.

paul vixie