Interpreting The Results
Below are three 802.11 (i.e. Wi-Fi) scenarios we'll consider:
1. Both the transmitter and receiver use a wired connection to the network. We connected our two machines to a wireless router that included 4 100-Mbps ports. The important point about this scenario is the path between the two nodes is wired on both sides. When we ran NetStress our results reported a throughput of a little over 7,750,460 bytes-per-second. A 100-Mbps port is capable of a theoretical, maximum throughput of 13,107,200 bytes-per-second (100 x 1024 x 1024 bits / 8 bits / byte). Considering that various factors (including the TCP protocol) contribute to an overhead that causes the actual throughput to deviate from the theoretical, then this type of result is reasonable and to be expected.
2. Either the transmitter or receiver communicates wirelessly -- the other uses a wired connection to the network. We setup a wireless network and configured the wireless router's Network Mode to use Wireless-G only. When we ran NetStress our results reported a throughput of approximately 2,500,000 bytes- per-second. So, how does this compare with the theoretical maximum of 54 Mbps? 54 Mbps is equivalent to 7,077,888 bytes-per-second (54 x 1024 x 1024 bits / 8 bits / byte). So, is our actual throughput (which is roughly 35% of the theoretical) reasonable? The short answer -- yes. The 54 Mbps is the theoretical maximum of what the hardware and medium are capable of. There are numerous factors that can contribute to the difference -- including network traffic, interference from other wireless devices and, equally important, the overhead of the 802.11 and TCP protocols. The throughput we measure doesn't take into account that every 802.11 packet includes additional bytes besides the data payload. Also, the TCP protocol takes a big chunk out of the performance (http://repository.cs.aueb.gr/getdoc.php?papercode=CPWR-18-2006). Taking into account throughput and coverage factors, others (http://www.atheros.com/pt/whitepapers/Methodology_Testing_WLAN_Chariot.pdf) have determined a theoretical maximum user-level performance for the 802.11 standards -- and for 802.11 g this turns out to be 24.4 Mbps (approximately 3,200,000 bytes / sec). Our measured throughput of 2,500,000 bytes / sec compares favorably with the theoretical maximum user-level throughput of 3,200,000 bytes / sec.
3. Both transmitter and receiver machines communicate wirelessly. A setup similar to (2) above, except both nodes use a wireless connection to the network. When we ran NetStress it reported a throughput rate of 1,300,000 bytes-per-second -- roughly one-half of that reported by scenario (2). The explanation is that each wireless "hop" takes its toll on the overall throughput -- so, this result is reasonable and expected.
Conclusion:
We've presented some of our results here to serve as a guideline to help you with interpreting yours. Your throughput benchmarking results will likely differ from ours but, in general, for similar setups we'd expect the results to be roughly comparable. That is, of course, unless there is something critical going on that is impacting the performance of your network.
Just to be clear, you should not use our results as a reference. The best reference values are those you record during the lifetime of your network. Only in that way will you know for sure whether things are working normally or whether a problem has arisen which now requires further investigation.
Copyright © 2010, Nuts About Nets, LLC