Subject: Re: Strange network performance
To: None <gij@jk.priv.no>
From: Kevin M. Lahey <kml@nas.nasa.gov>
List: port-i386
Date: 09/16/1998 10:41:41
In message <199809161615.SAA21511@ragnarok.si.sintef.no>,
Geir.I.Jensen@Runit.Sintef.No writes:
>My network performance is way below what a Linux box collogue of mine
>achieves (ok, so its not the same underlying hardware, but...)
>
>In normal use, the throughput is under 100KB/sec on a 100Mb channel!
>There must be something wrong somewhere. I wrote a script to test the
>performance for various packet sizes. The resulting graphs are located
>at (I didn't want to bloat the list): 
>
>	http://www.jk.priv.no/~gij/NetBSD/     (128.39.6.12)
>
>We can see a substantial drop in performance around packet sizes of
>100 bytes and 200 bytes. I assume this is a driver problem.

This sounds like two problems.  There are known problems with
the BSD stack for certain packet sizes that correspond to the
limits of mbuf sizes, and have to do with when the kernel decides
to switch from using two mbufs together to using a cluster mbuf
with external storage.  I think we've had an outstanding PR on
this for some time.

The rest of it sounds like driver weirdness, and I can't help you
there.  Having promiscuous mode on really shouldn't matter, should
it!?  Ack.

Good luck,

Kevin
kml@nas.nasa.gov