At work, we have, in our small group of researchers, 6 networked computers with varying specs connected by a gigabit switch.

Lately, it has occurred to us that the transfer speeds are very, very different among those computers. It would seem that a few of them are always slower than others, and that by a very large margin! I tried measuring network transfer speeds by doing a simple linux "scp" copy and found that while some of the computers transfer @ 11MB/s sustained rates, many of them do not get over ~800-900KB/s when transferring a 100MB test file!

What is going on here? I know that they're all connected to the exact same switch... are NICs so terribly different??...

This might be why I've never seen sustained speeds of more than 10% of Gigabit ethernet's maximum theoretical bandwidth, 125MB/s. Maybe that would require a dedicated high-end NIC like this one?....

yeah, nics do that. maybe it's packet collisions. btw, how's your professor's rig goin'?

Really? Wow, I didn't know nics could be so different. But it kind of makes sense. After all, onboard nics are typically not used for high-bandwidth tasks, I suppose...

About that rig: that's... oh man, that's just a horror story. We got all parts here, except the Tyan i5400PW motherboard, which is stalled at customs because of a federal police strike here in Brazil. They're striking for almost a month now. We're taking legal action to ensure that the motherboard gets delivered, but so far it hasn't helped.

Doesn't it sound like a bad joke? No, really: there's 16GB of FB-DIMM DDR2-800 and 2 Xeons X5472, plus two thermalright HR-01-X, JUST SITTING HERE DOING NOTHING for several weeks now. Isn't it revolting?

An IT horror story, really.

That's tragic. By the time you get it, it could be depreciated to a point where an entirely new system could cost the same as a motherboard.