Are all NIC created equal?

[Colin Morey]

hmm, well I had to raise an eyebrow at his use of a switch,.. I would have prefered him to do his testing on a cross-over connection, as a switch will always add latency into the equation.

One of these days I'll do some testing with 2x3com cards, and see what the max speed I can obtain on this network at home.

Colin

[spoogenet]

At least he used a switch rather than a hub. However all switches/hubs are not created equal either....

b

[dZeus]

ok, I just found the reason why I got the higher CPU utlisation that with my previous tests... I had ftp transfer set to 'ascii' in stead of 'binary'.

With binary, I get like 70Mbps, and 30% CPU utilisation on the Celeron 333, which is running Linux. (all test were transfering data from RAM to RAM over 20 meter CAT5 UTP and a 100Mb hub.

The Win2k workstation however uses 80% CPU util with the same transfer speed.. hmmmm... and that's a Coppermine @ 933Mhz as well... so I guess the Win2k default drivers aren't too good for the 3c905c-tx. I'll try the latest drivers soon.

[xortam]

These laboratory tests of NICs are interesting but I think you really have to see how they perform in real world scenarios where collisions occur and such.

[dZeus]

I just tested with the nebios/smb over tcp/ip protocoll in stead of ftp, and now I get like 72Mbps with 35 - 40% CPU on the Celeron 333 (linux) and 10 - 15% CPU utilisation on the Win2k workstation... so it seems that windows likes smb MUCH better than ftp, and linux doesn't really care (though ftp seems to be a bit better...)