It's easy:

On one CPE start the speed test in server mode (just check the radio button labeled "Server" and click on "Start"). The server will run until you stop it, even if you close the speed test window. On the second CPE the speed test needs to be started in client mode (selected by default). Type in the IP address of the first CPE in the "Server IP" field, then click on "Start".

That's all. You can choose to test the speed of unidirectional or bidirectional data transfers by selecting it in the "Direction" field. Similar you can define the number of parallel connections (1-20 in field "Parallel"), but the defaults are reasonable.

I was in the exact same situation a few weeks ago until i figured it out.

So assuming you have 2 CPEs, having assigned static IPs first one is for example 192.168.1.11 and the second 192.168.1.12.

You log in to the first one, for example 192.168.1.11 go to the speedtest tool page, click on the server radius button and select start (so as to the server on that side of the comms link). Then on another browser window / tab you log into the second CPE (i.e. 192.168.1.12) and on the speedtest tool page you leave the client radius button selected, input the first CPE's IP in the server IP (i.e. 192.168.1.11 for this example) and click start. That's all there is to it, however in order to interpret the results one has to really perform literally dozens of tests (especially in so-so situations due to either distance, or freshnel zone obstaces, or even maybe both) given the several combinations of channel width (5/10/20/20-40), channels, MCS and what have you...I've made a spreadsheet somewhere to keep a reference in my scenario, took a whole afternoon to weigh in about 20 different such scenarios to come up with the best combination..

It's important to note that unidirectional means one-way communication (i.e. from the server to the client / Rx for the client that is), whereas bidirectional is in fact simultaneous sending/receiving (Rx/Tx) which stresses the link a bit more and as a result you should expect one metric for the unidirectional test and two metrics (which adding together usually yield somewhat lower results than the one-way unidirectional one).

Hope this helps, now on with the tests :)

It's the number of parallel threads executing during the speed test. It roughly corresponds to simultaneous users creating traffic during normal use.