> While I do see your point, that would give me a 'false' sense of what
> my entire server could produce for throughput. Although I am
> interested in the raw speed that the NIC only can produce, I am more
> interested in the servers ability to deliver a consistent rate of
> transfer (hopefully above 200Mbit).
Well, you can stumble around trying to guess what the problem is, or you
can systematically attack the different possibilities, in order of
likeliness. It's up to you. : )
I'd start by testing the throughput of the network itself. Make sure
that nothing there is a limitation. That alone will narrow down the
possibilities by a huge amount.
Next, does your card support interrupt coalescing? I couldn't find
reference to that card having support, but I could be wrong. Trying to
run gigE adapters on a single-CPU machine without it can swamp the CPU in
interrupts, keeping it from being able to do it's job, whereas your
(assumed dual-CPU) Xeon machines can let one CPU handle the interrupts,
while another still does useful work. However, even without it, I would
still imagine that you'd get more then 200 mbits/second on your machine.
Since you're doing FTP, can your disk system keep up with 100+
megaBYTES/second? Can the disk system at the other end? There are a lot
of factors involved. Getting gigabit speeds at the *application* level
certainly isn't as easy as sticking in a gigE card and adjusting some
kernel parameters. On the other hand, you should still be getting more
than what you are seeing.
steve
|