North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Internet II is coming...

  • From: Paul J. Zawada
  • Date: Wed Oct 09 14:09:26 1996

At 11:06 AM 10/9/96 -0400, you wrote:
>
>"Dorian R. Kim" writes:
>> This sort of proposal, i.e. building a Higher Ed private network for
>> research, is in and of itself not such a bad thing.
>> 
>> The grow of Internet since NSFNet shut down has put serious strains on the
>> infrastructure that researchy folks used to use to do(and still do) their
>> various work on.
>
>You know, maybe I'm crazy but I rarely see the troubles that people
>mention so often.

The problems are real.  We (NCSA) have users that can ftp a 500MB-1GB file
to a site relatively close (Internet-topology-wise) but the the damn thing
times out when they try to ftp to it to us, two NSPs plus a NAP or private
interconnect away.  We're sure it's not our pipe to the Internet, since we
have a DS3 that _peaks_ at 50% utilization and usually hovers somewhere
around 10%-20%.

The problem is probably less the way the Internet has been built but the
dynamics of TCP.  The current Internet architecture only magnifies the
problem.  As Matt Mathis points out, the needs that Internet serves today
and the needs of high-end network users are orthogonal.

>
>When I'm going between my site and another site on the net, if both
>ends are unloaded, I typically get bandwidth equal to the smaller of
>the two pipes into the net. Its very rare that I don't get transfer
>times near the maximum expected, even when one of the pipes is
>attached to a mediocre provider. (Really bad providers are another
>story, but I luckily can usually convince my clients not to use them).

Have you tried this with two DS3 pipes at opposite ends of the Internet
(i.e. from two different NSPs at different ends of the US)?

>Seems to me that if the university researchers are sick of competing
>with the undergrads, either the university could get a fatter pipe, or
>they could priority queue the traffic from the researchers, and either
>way they would probably win. 

But they can't priority queue traffic coming back to them.  If the
undergrads' incoming traffic keeps their ACKs and return traffic from
getting back to them, they're still screwed.

>Even with all the well-publicized growing
>pains at the providers, I think the trouble is most likely at the end
>points, and not in the providers.

What they want is a fatter pipe with end-to-end priority queing.  (Give me
the list of ISPs providing this today or have announced it.  I want to talk
to them.)  This way the researchers can take advantage of the extra
bandwidth they're paying for and the undergrads can continue using what
they're using today.

--zawada

  
Paul J. Zawada, RCDD     | Senior Network Engineer
[email protected]     | National Center for Supercomputing Applications
+1 217 244 4728          | http://www.ncsa.uiuc.edu/People/zawada

- - - - - - - - - - - - - - - - -