On Wed, 27 Sep 2000, Lucio De Re wrote:

> On Wed, Sep 27, 2000 at 12:35:21PM +0100, Jasper Wallace wrote:
> > 
> > This might be a full-duplex/half-duplex negotiation issue - what did/does
> > your switch think the port is running at compared to what netbsd thinks
> > it's running at?
> > 
> I don't really know these things (I'm totally baffled about the
> switch, I can't figure out how it matches 10MHz segments to 100MHz
> segments if it's as unintelligent as I think), but when the two hosts
> were on the same hub, the problem was at its worst.
> 
	Do you have the card on autoselect? Could you try 
	'ifconfig <IF> <ADDR> 100baseTX' on the 100base hub to ensure it
	is not trying full duplex.

> I'm not sure how to check, we try to lean on others for our network
> wiring savvy.  As you can imagine, the results are less than
> satisfying.
> 
> But the problem has been exaggerated beyond acceptable limits by
> upgrading from 1.4 to 1.4.2, which strikes me as curable.
> 
> Note these odd FTP times for the 2MB SCO Unix kernel:
> 
> NetBSD 1.4.2 host on 100Mhz segment -> SCO host:	2.1 seconds
> SCO host -> NetBSD 1.4.2 host on 100MHz segment:	39 seconds
> NetBSD 1.4.2 host on 10Mhz segment -> SCO host:		2.1 seconds
> SCO host -> NetBSD 1.4.2 host on 10MHz segment:		19 seconds
> NetBSD 1.1 host on 10Mhz segment -> SCO host:		6.1 seconds
> SCO host -> NetBSD 1.1 host on 10MHz segment:		7.6 seconds
> 
> Of these, the last ones would be preferable :-)  Maybe I should route
> traffic through an intermediate host running an old version of NetBSD

	What do you get between the NetBSD 1.1 and NetBSD 1.4.2 hosts,
	and between the SCO host and another non NetBSD box?

                David/absolute
			       -- www.netbsd.org: A pmap for every occasion --