[ previous ] [ next ] [ threads ]
 From:  Michael David Joy <mdjoy at phy dot olemiss dot edu>
 To:  m0n0wall dash dev at lists dot m0n0 dot ch
 Subject:  Re: [m0n0wall-dev] Initial 1.6 Gigahertz AMD Opteron Benchmark
 Date:  Fri, 10 Sep 2004 09:07:32 -0500
> How big are your socket buffers?  Note that at Gb speeds, each additional
> millisecond of RTT needs an additional 125K of socket buffer to avoid
> being window-limited.  And of course RFC1323 window scaling needs to be
> enabled for this to do any good.  Even "short fat pipes" need big windows
> when they're sufficiently "fat". :-)

I'm not sure what my socket buffers are. The box is in a plain vanilla
m0n0 pc config. Whatever configuration the default cdrom image comes as
is what it's configured as. 

For the record, I'm not using jumbo frames (most of our network here is
jumbo fragmenting in nature), and I'm not sure whether the RFC1323 is on
by default or not. I assume it is on by default, since this flag is what
keeps some m0n0 users from going to www.netbsd.org due to a many year
old bug in the netbsd tcp layer.

So... without a real console, or much experience with BSD I'm not sure
where to look first or what to do to change the tcp buffers. If someone
wouldn't mind walking me through it, I'd glady take a look and tweak
some more and provide more benchmark numbers.

I assume that possibly linux has larger tcp buffers by default at least
in Fedora Core 2, partially accounting for the 55-60 MByte/s as apposed
to the m0n0 routed 40 MByte/s? Do remember that all 4 talking gigE
adapters are identical broadcom chips on a 64-bit PCI-X bus at 100 mghtz
(bandwidth is not an issue).

Anyone care to enlighten the BSD NewB here? (I.E.ME).