Subject: Re: TCP_MSS to 1460 in tcp.h or /etc/netstart?
To: Ken Hornstein <email@example.com>
From: Greg A. Woods <firstname.lastname@example.org>
Date: 10/27/1997 18:04:26
[ On Sat, October 25, 1997 at 23:46:27 (-0400), Ken Hornstein wrote: ]
> Subject: Re: TCP_MSS to 1460 in tcp.h or /etc/netstart?
> Wolfgang Rupprecht <email@example.com> wrote:
> > (It seems to be a common misconseption for folks setting up their
> > SLIP/PPP links to request a really small MTU in an attempt to get a
> > better interactive response. In the absense of TOS queuing, I don't
> > see how it matters much if an interactive packet is stuck behind 3 500
> > byte packets or one 1500 byte one.)
> With all due respect ... I've tried it both ways, and it _does_ make
> a difference for me. But as always, YMMV.
Ken I'd be interested in knowing if you've monitored for overruns and
other errors during your tests.
If there are any errors at all I've found that a lower MTU can make a
big difference not only to interactive response but of course to overall
throughput too if you're using a machine that's only marginally fast
enough to run its own ports.
Of course the serial driver could also do better and try and stem the
flow when things start getting out of hand and thus prevent the overruns
in the first place. Properly functioning hardware flow control should
still allow faster recovery than retransmission. There's really no
excuse for any overruns ever at all on anything more modern than a 286
so long as your hardware flow control is properly connected and working.
Greg A. Woods
+1 416 443-1734 VE3TCP robohack!woods
Planix, Inc. <firstname.lastname@example.org>; Secrets Of The Weird <email@example.com>