Follow the Bouncing Throughput
That being said, let's turn to the first issue that I found. While I was running some of my early tests, I kept seeing a noticeable difference in maximum throughput. At first I thought it was due to some of the interference tests that I was running, but then I decided to do a special test. I made sure that the only wireless equipment powered up was the WBR-G54 and Buffalo WLI-CB-G54 client. I then ran the NetIQ Chariot throughput.scr script for about 20 minutes, modifying only the file_size parameter to use a 200,000 Byte file size, instead of the default 100,000 Bytes for each transfer. Figure 10 shows the result.
Figure 10: Long throughput test
This proved that I wasn't imagining the max throughput difference, and further showed a definite cyclical pattern. A similar pattern can also be seen with a cycle time of somewhere between 30 and 60 seconds. So Issue #1 is that there is work needed, probably in the transmit-rate auto ranging routine. When I presented my findings to Buffalo Tech, they confirmed the issue, and a few days later sent the plot shown in Figure 11.
Figure 11: Long throughput test - new firmware
The plot was generated with tweaked, unreleased firmware and although still shows a cyclical throughput variation, the variation has been reduced. So Buffalo (and presumably Broadcom) is on the problem, and says they'll be issuing a firmware update in mid-January. However, all the following results should be viewed with this issue in mind, because it can be seen in pretty much every test.