[c-nsp] FTP Throughput

[Jared Mauch]

On Nov 29, 2011, at 3:24 PM, Joseph Mays wrote:

> Running tests on FTP throughput from a windows ftp client across two T3 hops to an ftp server running on FreeBSD unix. Pretty much all the bandwidth on both T3's is available. Total latency averages about 3ms. The customer on the end of the t3 is complaining that they can't get faster than 600KB per second anywhere. I get about 1000KB (8mbps) on a file transfer. I can start multiple file transfers, simultaneously, all top out at about that speed. The customer is demanding to know why they can't transfer files at, say, 40mbps. I am assuming the answer is something to do with TCP window size, but how do I prove that?

I would take a moment and setup jperf (win) and iperf (*nix)

test with those and identify any network elements that may be a problem.

Also, the "windows" systems have generally limited themselves to slower speeds depending on the version.  (Jim Gettys has identified this as part of his 'buffer bloat' activities).

Hopefully you can find out it's a simple duplex mismatch someplace.  make sure you check that in detail as it's quite common to cause the network to be actually be slow.

do both UDP and TCP tests.  I've also seen cases where there's a midpoint in a layer-2 domain where it's actually 100m or (back in the late 90's 10mbit) that was the central location between two 100m switch segments.  This obviously had some side-effects.

- Jared