> My personal experience is that on two IBM 305 gigabit Px2GHz servers
> using something like
> dd if=/dev/zero | nc slave 1234 on master
> and nc -l -p 1234 | dd of=/dev/null on slave
> so no hard disk bottleneck
> We had not more than 60Mb/sec with a 100% CPU utilization (35Mb sec on
> real disk access) and I'm sure these servers has PCI66 at 64bits called
> XPCI.
>
> So bottleneck is now CPU !
That's totally interesting. Nice little experiment ;-).
(I've generally noticed something similar with the servers on the
1Gb/s fiber-optic lan where i work. the bottleneck there seems to
be the circuitry within the computers themselves; the resistance
of the electrons on copper is far greater than the photons shooting
through dark fiber, obviously.)
lion-O
--
''Wisdom is the companion of patience''
,-~~-.___.
/ | ' \
( ) 0
\_/-, ,----'
==== //
/ \-'~; /~~~(O)
/ __/~| / |
=( _____| (_________|
Brett Ryan Campbell
Systems Administrator, CAD Research Center
Cal Poly State University, San Luis Obispo, CA 93407
http://www.cadrc.calpoly.edu/frameset_content/content_about_us.html
--
[EMAIL PROTECTED] mailing list
