We have a network product that runs on Linux and needs to be tested in reduced bandwidth networks.
We have been able to use 'tc' (traffic control) in Linux and dummynet in FreeBSD to inject delay (latency) with predictable behavior.
But when we try to throttle the bandwidth with either of these, it does not seem to have much effect.
Here is an example of trying to use 'tc' to set the bandwidth to 512Kbps:
tc qdisc change dev eth0 root tbf rate 512kbit burst 1540
But when we run 'scp' over that connection we see speeds that are way higher than we expect.
Here are results of setting various bandwidths in both dummynet and tc. You can see that it has SOME effect but not very much.
BW _______DUMMYNET__________ __________ TC __________ 10Mbps 1m,3.489s 4.0MB/s 1m,2.732s 4.1MB/s 8Mbps 1m,3.461s 4.0MB/ 1m,2.728s 4.1MB/s 6Mbps 1m,3.905s 4.0MB/s 1m,2.710s 4.0MB/s 1Mbps 1m, 3.261s 4.0MB/s 1m,2.680s 4.0MB/s 512Kbps 1m,26.175s 3.0MB/s 1m 25.409s 3.0MB/s 256Kbps 2m,39.079s 1.6MB/s 2m 53.004s 1.5MB/s 128Kbps 5m, 3.501s 852.7KB/s 5m, 50.491s 739.7KB/s 64Kbps 9m,48.907s 463.3KB/s 9m,53.105s 436.5KB/s
I don't have the syntax we've used for DUMMYNET handy but I'll post it later.
Best Answer
There is a question on serverfault that I answered with a more extensive example of using tc and iptables to do traffic limiting. This example comes from a setup we have on our mirror server and it works very well.