I have, as shown in the visio diagram, a Cisco 2924 XL switch that connects my entire network together. I have about 15 machines that may be connected at any one time. Some of these are Unix, Linux, NT, Win9x, laptops, PC's, Server's, etc.

Since all but about 3 or 4 have 10/100 cards in them, and they are all good cards, like the dual port SMC or the 3com 905x series adapters, I would like to see them all running at 100Mb full duplex. However, that is not always the case.

For example, my Red-Hat Linux 5.2 machine will absolutely NOT connect at anything other than 10Mb Half Duplex. Most of the others are ok, although my Solaris always wants half duplex too.

When I recently moved my "data center" to my new address, I had to layout all the machines again, so I decided to do some switch control while I was at it. I went into the Cisco Virtual Switch Manager, and locked all ports to their maximum capabilities, and turned OFF autonegotiation. I thought I was doing a good thing. Even though only two machines were not negotiating properly (as mentioned above), I thought I would simply make sure that the entire group always would.

Well, the short of is that my backups went from around 6 hours for 12 Gb of data to 44 hours for the same. My NTworkstation machine alone went from 9 minutes to over 2 hours 45 minutes? All this because I am just FORCING the 100Mb/Full Duplex, even though they already connect at this speed. So I went back to autonegotiation for all machines now and everything works as it should.

I must say however that it appears counter - intuitive to me.

NOTE: I will offer this bit of advise, based on my own findings - use Netscape 4.6 before using IE5 for the Virtual Switch Manager. On my system (ntws), even after 5 minutes, IE5 would not finish drawing the main screen whereas Netscape was done in 5 seconds !!