Different speeds

what can be the problem with these results:
link test 120/50Mbit
speedtest 50/50Mbit
fast.com 50Mbit
downloading iso of debian 99Mbit
torrent 110Mbit
playing youtube video 4k 50-70Mbit - playing is ok

when i test speed on speedtests, is download speed max 50Mbit.


I think the link test is UDP (torrent, file download too), but the speedtest is TCP, which will always be lower because it requires a response. Cambium marketers came up with the idea of presenting UDP because it seems to have a higher speed. ::

From what I see, there is nothing wrong with the equipment.

first, your tests are not just your network except the link test which is just AP to SM, you can not be held responsible for data speeds after it leaves your network (though clients will try to! make sure your contract with them specifies this and that service is “best effort” so you cant be held for lost packets or congestion periods), but you can be held responsible for not having enough bandwidth to support those speeds for at least 10% of the clients with that rate plan.(most countries have this implemented some how to protect consumers, it may be more than 10% in your country)

second, you’ve proven the link as capable to the plan speed, now prove your network isnt the problem:
setup an iperf server at the gateway of your network (on the uplink side of your gateway router is best just set it to only allow connections from your IP range), then run Jperf from your laptop at the clients SM (no through their wifi router unless you want to fight with explaining slower speeds due to wifi sharing)

third, perform jperf tests for both tcp with a window size of 4000 and for udp. How you setup the test will give you many ways of looking at the results. dont forget to disable your windows firewall!

Fourth, cover your butt!
Add a clause to your client contract that states that testing to test sites outside of your network is not a valid test of the client’s connection through the network and shall not be used as evidence against the network.
Also add to all your marketing information the magic words “up to” before your advertised plan rates. This means that legally the rate may be less at times depending on many factors, just make sure your network can handle the bandwidth required 90% of the time! MRTG or Cacti is your friend here.
Implement a Fair Access Plan and publicize it and add it to the customer contract. In this plan, cover your butt by stating clients may be slowed down during internet congestion (not network! this is important). Get your lawyer to verify that it doesn’t break any laws in your country.

1 Like

Dear David

I have question in a situation. where I have SMs connected to a Base radio that has upto 10other SMs connected to it. Is it possible to carry out an iperf test on an SM that is connected to Base radio that has other SMs connected to it?

the epmp wireless link speed test IS iperf so yes.

the way the frame scheduler works in the AP makes this a shared transparent bridge so it will not affect iPerf, but iPerf can see the flow stops if your tcp window is too small hence the window size of at least 4000.

1 Like

Most of speed test (Ookla) servers are not implemented for long fat networks. Some are, you need to find the right server to test your speed against since you cannot tune their servers. You can set up your own and tune the TCP congestion control algoritm to be acording with a wireless network . Speed.ui.com works great, open-speedtest.com works great. Fast.com and Ookla (default algorithm CUBIC which is loss based) perform really bad specially when your AP starts to do its job with a fairly number of customers on it.