New MAE-EAST

Tue Nov 11 13:04:54 UTC 1997

"Kent W. England" <kwe at 6SigmaNets.com> writes:

> There is no point agreeing with the Big Backbone Network Engineers that the
> MAEs suck. It is in their best interest that the MAEs
> suck,

Hm.  Well, it depends on how deep you want to get into
conspiracy theories, of course.

If there is a way to sell normalized services over an IXP
such that the costs-versus-revenue split is not
significantly worse than offering normal services over
non-IXP technology, then there is no reason to dislike
IXPs.  There is technology to offer normalized services
now, and apparently it is being put to some use.  How much
this is thought of depends on how holistically you want to
view organizational financial structure and cash flow.

A facilities based telco that is at an IXP that is not
operated by itself is less likely to be thrilled by the
thought of relatively small Internet access fees available
there in comparison to the potentially very lucrative
business that can be obtained through bundling Internet
access as sugar in securing a more comprehensive account.

(Hi, we're a telco.  Buy our long distance and use our WAN
outsourcing services and we will give you nearly free
Internet connectivity.  -- This is difficult to do at an
IXP...)

On the other hand, an organization that is running a large
IXP and can do so without losing money -- for instance,
when they are provided with a captive and unwilling market
thanks to government pressure -- probably will be very
fond of them, particularly if the infrastructure being
paid for by the IXP participants can be turned into an
aggregation point for their own Internet service offerings
(and possibly, in the case of LECs who run IXPs, bundled
in with other services like inter-campus WANs or even VPN
telephony).

(Hi.  Welcome to our ATM switch.  Did you know that we can
make some VCs between you and University of XYZ, not to
mention that we can offer a whole range of unregulated
services thanks to this wonderful new NAP technology.)

The thing about Internet Engineers is even the most evil
greedy bastardlike ones mostly seem to want the Internet
to work.  Relying on IXPs which are bursting at the seams
technologically and physically seems really dangerous.

That some exchanges (I note you say MAEs) suck is a
side-effect of their popularity.  Keeping that popularity
from exposing scaling problems beyond the IXPs is an
intelligent design goal, which also may have convenient
financial implications.

> the CIX is crippled,

The CIX ceased to have any real function when ANS CO+RE captiulated.

Now that Rick Adams has finally eaten Al Weis's lunch, the
continued existence of the CIX is almost a joke.  Sorry,
Bob and John.

> you aren't bugging them to plug into a high perf exchange, and
> that you, the little ISP, go out of business soon.

Um, interesting theory.  Given some statistics on where
traffic loads are and who seems to present what amount of
aggregate traffic in the USA, I am not sure it's really
tenable though.

> THEY have private interconnects which you can't join.

How do you "join" a point-to-point circuit?

My position on these private interconnects is that each
party is offering some degree of connectivity and that
normal business negotiations on the price of those
services determines who pays whom what amount, if a deal
is to be made at all.  This is entirely like a negotiation
on pricing done between any two entities on the Internet.
Peering, 102: it's exactly the same as any other deal on
connectivity.  (cf. Vadim Antonov's question two years and
change ago, "does anyone ever actually pay list prices?")

Most of these private point-to-point circuits are
negotiated in circuit pairs, with each party paying for
one out of every two circuits.  Long discussions sometimes
happen about who should pay for which of a pair of
circuits and whether there should be some further
consideration (financial or otherwise) even within a
contractual framework that is geared to make this sort of
thing straightforward.

> Find a co-lo where you can cross-connect without being
> robbed or build your own NAP, just don't use
> DEC-designed Gigaswitches and FDDI. Use full duplex 100
> Mbps Ethernet switch or find an old Fore switch cheap.

This is good advice except that there are MTU implications
wrt Ethernet that need considering.  Personally I am
hoping that people start fixing little things like the old
decision to assign low default MTUs and MSSes to remote
things.  Something useful to consider is that there is no
difference between a properly designed router and a
properly designed switch, and that given a
next-hop-resolution scheme (tunnels, tag switching, you
name it) a router acting as a switch (a "srouter") only
needs to know about its immediate adjacencies in order for
connectivity to work.

	Sean.