On Sat, 6 Jan 2007 15:45:19 -0700
"Andy Ruhl" <acruhl@gmail.com> wrote:

> On 1/6/07, Steven M. Bellovin <smb@cs.columbia.edu> wrote:
> > I just picked up a Linksys re(4) board, in preparation to
> > converting my house to gigabit networking.  I've noticed two
> > problems with the board thus far.
> >
> > First -- it's much slower on transmit than receive.  It can receive
> > at at least 420K bps, but only transmits at 200K bps.
> >
> > Second -- it takes a *very* long time to autonegotiate with a
> > (Netgear) GigE switch, so much so that I had to add !sleep 15 to my
> > ifconfig.re0 file.  I'm not sure how I'd do that for a dhclient
> > host.
> >
> > I've tested this on a fast (dual core, 2.13Ghz) machine running
> > 4.0beta2 and a slower (1.5Ghz Athlon) one running 3.1.  Both
> > problems occurred in both systems, though I think the speed numbers
> > were (understandably) a bit better on the faster machine.  I can
> > get a modest performance boost by turning on checksum offloading,
> > but the difference between send and receive is still a factor of 2.
> >
> > So -- is this a problem with the board, the chipset, or the
> > driver?  If this is a hardware limitation, what gigE boards should
> > I be buying?
> 
> I bought a pair of cheapie cards off ebay that had the Intel chip (wm
> driver) and a Marvell PHY and they don't do what you're describing.
> 
I'm having enough problems with wm cards on some other machines that
I'm leary of that path.  (My laptop won't autonegotiate 10baseT, and
won't come out of 10baseT speed (as opposed to setting) if I set it
manually.  A server I have won't autonegotiate 100baseT, though thus
far it seems to handle GigE properly.  But that's the subject of another
post for another day...)


		--Steve Bellovin, http://www.cs.columbia.edu/~smb