We often get reports of users testing throughput using software based utilities like iPerf who complain that they are not getting the expected throughput. Unfortunately, we've seen many problems with testing the throughput with iPerf.
It seems that Iperf and others have a very hard time actually controlling the IFG (inter frame gap) to actually send a set throughput less than the interface capacity. It works great when you're testing a medium that is the same as the interface on the computer, but if you have a GigE interface and the radio can only do ~375 Mbps, iPerf will send GigE for some fraction of a second, then pause and do nothing, then send GigE, then pause, then send GigE, and pause. Unfortunately this causes the buffers to fill up and overflow which causes Iperf to report a low number. If you set your interface to 100 Mbps instead of 1000 you may see much better results. Normal TCP windowing solves this buffering problem.
Larger buffers to combat this problem would result in increased latency and jitter which would cause problems with live traffic at the cost of improved testing which is typically only done once. This behavior is not apparent with live traffic. The buffer management was improved in recent versions, so it is suggested that you upgrade to the latest firmware before doing any Ethernet testing.
I know it is hard to trust it, but because these are full duplex radios, they actually achieve the rated performance - or they just don't work. There really isn't anything in-between. With ACM, the link could shift to a different modulation that would carry less throughput, but again, it would be the advertised throughput for that speed. As long as the RF performance looks good, you should not experience any problems with throughput.
The best way to check the throughput is to watch the RF utilization in status port or to use a RFC 2544 test from a dedicated hardware based Ethernet tester, like Exfo, Xena, SmartBits, Fluke, etc.
Another way to test it if you can get Iperf to do it is to push UDP packets at full capacity and measure what is actually coming out the other side. I don't recall if this can be done with iPerf.
Comments
0 comments
Please sign in to leave a comment.