Date: Thu, 29 Jun 1995 15:57:07 +0300 From: Heikki Suonsivu <hsu@cs.hut.fi> To: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> Cc: freebsd-hackers@freefall.cdrom.com Subject: Re: FreeBSD as a router Message-ID: <199506291257.PAA00901@shadows.cs.hut.fi> In-Reply-To: "Rodney W. Grimes"'s message of 23 Jun 1995 05:02:59 %2B0300
next in thread | raw e-mail | index | archive | help
From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com>
> A bit of extra delay shouldn't affect throughput, especially with TCP large
> window support, or does it?
It won't effect long TCP streams, but it sure as heck raises cane with
UDP traffic (NFS for example) which has no window and must wait for
round trip times for packet acks.
Not doing windowing or similar technique means broken protocol, if it is
supposed to give high bandwidth on anything but a local network. And at
least to me it would seem stupid to design protocols which are supposed to
work reasonably in local networks only. The more so as network speeds
increase. NFS is an example of a bad design.
As what comes to latency, I can't see any considerable difference in delays
between ciscos and FreeBSD routers (both seem to add about 1ms per router).
In comparison, over atlantic 1ms per router would make about 1% of the
delay, all the rest being light of speed and the phase of the moon. Not
much to be done by latency improvements?
--
Heikki Suonsivu, T{ysikuu 10 C 83/02210 Espoo/FINLAND,
hsu@cs.hut.fi home +358-0-8031121 work -4513377 fax -4555276 riippu SN
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199506291257.PAA00901>