3000 Low TCP Throughput

We are installing the first 3000 (AP & stations) but I can't understand why there is a slower TCP band than UDP
Ap has 40Mhz size, with 75/25 duty cycles.
The cambium test shows 220/35 Mbps, but when tested in multi-session TCP mode, the download reaches less than half, while the upload is at maximum (80/35Mbps)
Everything makes the situation seem clean. The modulations always remain on the last ones (the best ones) and the retransmissions are very low percentage

Has anyone else happened?


Stations Force300-16 on ePMP3000

Test Cambium UDP vs Test Mikrotik TCP

Modulation statistics Downlink

What model of Mikrotik are you using and what's the CPU usage look like when you're running this test?

I try with CCR to CCR, CCR to hAP-ac, 3011 to hex-s

Also real test with pc and speedtest

All test show same results

Please try 3011 to 3011 or CCR to CCR - with latest MT firmware.

After 6.44, MT started supporting multicore use for Bandwidth Test. Before that, no matter how many CPUs are present on RB, only one could be used for MT BW Test. We have tested with RB3011 to RB3011, also with RB4011 to RB4011, to see 230+ Mbps on TCP DL, even with TCP bidirectional tests.

Also, please try increasing number of 'Connection Counts" to 100.

I try with CCR and also 3011 on both side (version 6.44.3).

CPU of Mikrotik are not 100% when running.

Results are bad anyway

100 TCP connections improves by 20 mbps, but it is an unrealistic case.
When a customer makes a speedtest, this, doesn't open all those connections.

So finally, customer can't reach 100Mbps of throughput (his plan on my case)

Try to lock max uplink modulation for this user to MCS6 and check if it's better. If not, try MCS5.

I know you'll see lower throughput, but in some situations it can really help on TCP throughput and jitter.

Try it and let me know

Problem isn’t MT BT. Problem is TCP throughput on Cambium. Internal test shows 3-4 times better results than you can really get on any BT.
We have the same problem.

> Try to lock max uplink modulation for this user to MCS6 and check if it's better..
Hi giuseppe4, I confirm that with MCS6 in uplink customer reach 85Mbps (~ +20Mbps), a bit lower with MCS5

Better, but still far from 200mbps

> Problem isn't MT BT. Problem is TCP throughput on Cambium..

Hi pospanko, I am in fact very unhappy with the performance, with what these cells cost

Yes, it's a small workaround, but Cambium has to fix TCP throughput in some interfered situations.

I saw better performances with less retransmission, but I had similar results in other situations with more retransmission due to interferences.

AC platform is still unmature. Still lots of work to fix it up. Too soon Cambium released it. We have 60pcs in stock waiting for fixes to start deployment, for more then a month and half now…
Hope they’ll fix it soon or I’ll return everything to my distributer.
We are new to Cambium but their reputation was that their equipment work out of the box. And work well. Now they shooted in their own leg with to early distribution. If every one waited 4 years for AC products, we could wait 6 more month to get usable product.
Just my 2 cents…


I now have some 3000 deployed and we are having the issue with the low trhoughtput in the realworld. it keeps sppeds very low even with the SNR at 35~38. 

Hi Edgar - What firmware version are you using on AP and SM's.  Also, are you NATing on the SM's, or using Bridging?

We are having the same issues along with other issues.  Cambium is kind of quiet on the situation.

We too are experiencing the same problem with epmp 3000 on 80 MHz channels.

Very high UDP throughput (around 300 Mbps) and very low TCP speed (60-70 Mbps).

Still waiting for answers from Cambium. 

We see the same problem

Very good post, I have a network 100% rocket AC, I have to make an exit for the fiber and one of them that I am thinking and cambium but I need the radios to work perfectly, to compete I will wait for news in the forum and come the LTU from ubiquit to PMP with the airfiber 5x HD and compare the technologies, the better behave I will invest or run to fiber soon and end this dilemma.

OK, I've just upgraded my ePMP3000's and Force300's to Version 4.4 (RC30) and I've done some throughput tests.  What I've done is to have a MikroTik RB4011 connected to the AP's ethernet, and another RB4011 connected to the SM's ethernet - so that I can test the real throughput that the ePMP Link is capable of carrying, using the MikroTik's test.  The RB4011's are running 6.44.3

So - first testing in 40Mhz channel widths with 75%/25% ratios, and testing in TCP, using the 'Both' (Bi-Directional at the same time) test.  This test was 249 Megabit download and 54 Megabit upload - so an aggregate of 303 Megabit per second in a 40 Mhz channel.

I also tested in 80 Mhz channel widths - although it clearly says "Beta Quality" on the 80Mhz modes.  For this test I wanted to the one-way 'upload only' and then the 'download only' and then to also do a 'both' Bidirectional test.  In the following image, this is showing the AP's throughput graph - showing how much sustained traffic it was carrying, while the Mikrotik 4011's were doing their tests through this link.

This is 80 Mhz channel with 75%/25% ratios.

The Upload only test was 191 mbps (25% ratio)
The Download only test was 536 mbps (75% ratio)
The 'Both' Bi-directional test was 532 + 174 = 706 mbps aggregate


Nice testing ninedd! I'm happy to see RC30 is doing so well!!!

300mbps TCP traffic on a 40MHz channel is great! I would have expected something a bit lower actually, like around 260mbps aggregate.

I was a bit puzzled by some of the results too - I was surprised on the 191 Mbit upload speed in 80 Mhz too, but that's what MikroTik was able to put through the link. That was way more than I expected.

I'll have to sit down and do some really comprehensive tests, and test everything the same in 20, 40 and 80 Mhz - both UDP and TCP - and also doing receive, send, and both tests.  Also, I'd like to do them in TDD and also TDD PTP with only a single SM and see the difference.

BUT - that's a lot of tests... like 36 different combinations to test right there.  AND, I need to test the PTP550's with v4.4 also - so that's quadruples the number of tests there again - 100's of combinations.

So, by the time I'm done testing every possible combination... the ePMP4000 will be out. :D 

1 Like