Has anyone noticed that bandwidth test running tcp pegs the cpu out. I setup a ptp with 2 411a's in the shop to test nstreme and different settings. I at first thought it was something with nstreme but it happens with nstreme disabled. Anyone else seeing this??

thats pretty normal. its a lot of work to generate packets compared to just routing them. normally a true bandwidth test uses 3 machines, with the one in the middle being tested for throughput and the 2 ends running the testing.

Is it safe to assume that it is not a true measure of tcp throughput then?? Or do the test come out similar using iperf or some other test.

I see sometimes 5-10mb throughput improvement over built in bandwidth test by running through the devices instead of between them. If you use the built-in test just assume they are real world worst-case numbers. What you get should be better through the link.

Scott

running a link test requires 4 units. One server and one client and two units in between. They maintain the link to be measured. Their cpu's only deal with the traffic handling now, not with producing and comparing packets...