NAP/ISP Saturation WAS: Re: Exchanges that matter...
salo at msc.edu
Fri Dec 20 15:42:49 UTC 1996
> From: Jim Van Baalen <vansax at atmnet.net>
> Subject: Re: NAP/ISP Saturation WAS: Re: Exchanges that matter...
> Date: Fri, 20 Dec 1996 03:57:47 -0800 (PST)
> ... Why does everybody seem to be so
> sold on Gigaswitch based Xchange points? ...
A few random thoughts:
o I believe that the DEC GigaSwitch has been demonstrated,
under very heavy production loads, to be far superior to any
other available technology. Technology superior to the
GigaSwitch might exist, but I don't believe anyone has
demonstrated that technology under heavy, 24x7 load.
The alternative technology needs someone to bet their
exchange on putting 200-400 mbps of continuous Internet
traffic across the new technology.
o When the GigaSwitch was first selected for an exchange,
alternative technology was not as well developed. ATM switches
still had, (with somewhere between zero and two exceptions),
small buffers. A lot of ATM equipment was simply bad, at least
for wide-area applications. The fact that the Cisco AIP card
did not have a DS-3 interface also hurt, (fairly seriously,
it turns out).
o Note that if an exchange bets on ATM, its fate also depends upon
the quality of the ATM interfaces in the attached routers.
In particular, the Cisco AIP card becomes a critical component
in the exchange. Before I placed too heavy of a bet on an
ATM exchange, I would want to see a good performance analysis
of the Cisco AIP card and the ATM interfaces for any other
router which might attach. If the router ATM interfaces are
not as good as the router FDDI interfaces, an ATM exchange,
as a whole, is likely to be less successful than an FDDI
exchange. By the way, I would assume that router vendors have
spent more resources on improving the performance of FDDI
interfaces than ATM interfaces.
> Based on membership and traffic it
> appears that there is still a stigma associated with Xchanges (PBnap and AADS
> for example) that have chosen different architectures. ...
I think several unfortunate events plagued the PacBell and AADS exchanges.
o Because it was not clear that a Cisco DS-3 AIP card would be
available in the required timeframe, a decision was made to
use ATM DSUs and the Cisco HSSI interface, (a decision in which
I played a modest role).
o It turns out that the selected ATM DSU had implementation problems
which took a while to work out. ATM DSUs in general, I believe,
also have the problem that they simply aren't designed for the
sort of load that is experienced at an Internet exchange. The
Cisco HSSI card might also be weaker than the Cisco AIP card.
o It is possible that the two exchanges remained committed to
an ADSU solution longer than they should have, particularly
when it became apparent that the Cisco DS-3 AIP card was available
(something I had no role in).
So, history may play a bit of a role in the current position of the
PacBell and AADS exchanges. On the other hand, it appears that the
Internet operational folks are unlikely to try anything as new and
as unproven as ATM until they have no other choice.
I suspect that the next time ATM products and standards are called
upon to carry production Internet traffic, they will be ready to
support a heavy duty Internet load. I think ATM products and standards
are [perhaps almost, perhaps just] ready to support the traffic loads
experienced at Internet exchanges. It would be nice to see a heavily
loaded ATM exchange or perhaps a mixed ATM/FDDI exchange.
More information about the NANOG