Duplex Question

[Bob Ferro]

We have a rp3440 test server attached to a Cisco 3500XL Catalyst switch. The port is set to 100/full/no_auto. The rp3440 /etc/rc.config.d/hpigelanconf file is set to auto_on. I read in the forum to turn off auto_on in the conf. If I do, the server runs slow. I am testing it by doing a Data Protector (omnir) restore from the production server (rp5470) to the rp3440 server. If I do a lanadmin -X 100FD, the byte counts slows down. I monitor the restore using the DP GUI. Any ideas?

[Pete Randall]

The generally accepted rule of thumb says to set both the switch port and the NIC to 100FD for best results. If doing so does not give you a satisfactory result, then I wouldn't do it.


Pete

[A. Clay Stephenson]

Never, ever have the switch port and the port on the host set to different values. Both must be set to auto-negotiate or both must be hard set to a fixed value. Your configuration is all but guaranteed to result in exactly the situation you observe. Except for gigabit connections (which must be set to auto-negotiate), I still prefer to hard set the values but they must be hard set to match on both ends of the connection. Auto-negotiation works well almost always but there is a problem that can confuse the negotiation protocol at certain cable lengths (off the top of my head, somewhere in the 35-41m range) and for that reason I prefer to hard set them.

I actually remember finding a reference to this (and it's not confined to HP equipment):

Ah, here we go:

http://docs.hp.com/en/601/autonegotiation.pdf

Look at page 8 --- and I did correctly remember the cable length range.