North American Network Operators Group

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

Re: Global Crossing Network Problem.

  • From: Joe Abley
  • Date: Wed Jun 28 16:05:19 2000

On Tue, 27 Jun 2000, Alex wrote:

> Hmm..
> err, throughput is unaffected by distance. 

The bandwidth-delay product on the 200 mile OC12 is
622E6 * 2*(200*5280E-9) == 1313007 bits == 160 kbyte
(assuming symmetric route over the fibre, doubled to
give RTT).

Most TCP implementations ship and run with default
maximum transmit/receive windows that are much smaller
than this (e.g. a default 16kbyte window would give a
maximum theoretical throughput on a single, monstrous,
pipe-filling TCP session of 62,060,606 bit/s on the
200 mile OC12, but 124,121,212,121 bit/s on the .1
mile OC12 -- where it would fill the pipe at 622M
(window size/RTT, RTT = 2*{0.1, 200}*5280E-9).

Hence in the (naturally entirely common, see it every
day) case where you have colocated, TCP-speaking
devices at each end of your OC12 that are capable of
shifting data at those kinds of speeds and which have
their maximum TCP windows set way too small, and
ignoring all delays due to queueing, TTL-decrementing
and checksum calculating, encapsulation overhead and
transmission framing delays, the .1 mile OC12 is about
560Mbit/s faster than the 200 mile OC12.

You should say "aggregate throughput for a typically
large number of concurrent sessions" if you want to
avoid tempting irritating pedants out of their caves :)

> On Tue, 27 Jun 2000 [email protected] wrote:
> > On Tue, 27 Jun 2000 19:56:08 EDT, Alex said:
> > > is a .1 mile OC12 faster than a 200 mile OC12?
> > 
> > Yes, by 200 * 5,280 nanoseconds.  Grace Hopper said so. ;)