TCP bottleneck from Netonix to A5c to C5c's


#1

We’re running into a bottleneck issue and haven’t been able to root it out yet. We’re testing Mimosa gear on an access point with 4 C5c test clients and we aren’t seeing speeds on the CPE ends higher than 45-55 Mbps, despite the PHY being 300-400 and the spectrum being pretty clean. This A5c is being fed through a B24 link that can only do about 220 Mbps total TCP throughput with a PHY of 1170/1170. They’re all plugged into a Netonix WS-12-250-DC.

The Netonix is showing zero errors on the ports, and I’m not getting any dropped packets when I bandwidth test from Mikrotik hEX behind a CPE to a hEX plugged into the Netonix. Though, the latency is kinda high at ~25ms. CPEs are less than 1mi away, gains of -39 to -45 dbm. Flow control is set to “Both”. Using SRS mode.

On the other hand our Cambium AP plugged into the same switch hits 85Mbps on the CPE.

Any idea what could be causing this and what I can try to address?


B24 link only ~200Mbps
#2

Hi @Josh15,

I just responded to you via Facebook, but I’ll paste my response here as well:

I’ve personally never had an issue with Flow Control on the Netonix, but I know a lot of folks turn that off.

As for across-the-board throughput issues, I would start at the building you’re backhauling from and work my way back through the network from there.

If I were in your shoes, I would start with the switch port/router port your network is plugged in to. I’m not sure if you’re using Netonix on the building, but I think you are. So I’d bypass that Netonix and run a speed test. If that works fine, I’d put the Netonix back in place and test from the Ethernet port the B24 is plugged in to. If that works, move to your office and test from there.

You’ll need to do this step by step to find the issue. Once you do find what’s causing the issue, then you can work at finding why that particular piece is having issues.

As for your latency on the client side, you will see 20+ms latency while using SRS mode. Wifi mode you’ll have much less latency, but you’ll start having more issues as you start to fill up your APs. Once you start having 15 clients on an AP, you’ll start having CPEs try and contend with one another for air time. This is when you should definitely move over to SRS mode, since everyone has their own timeslots.


#3

Thanks for the troubleshoot walkthrough Dustin. We’ll give that a go and try to isolate the issue a bit further.

Just to be sure I’m not chasing ghosts, in your experience if the PHY is 1170, should I be expecting a TCP throughput higher than 200 Mbps? I’m not crazy for thinking that’s low, right?


#4

How wide of a channel are you using on your A5? Would you mind posting your channel config?

I have tested 40MHz on an A5 and pushed around 200 Mbps to a client, (Mikrotik RouterBOARD 750G for the testing) but we had to switch back to 20MHz because of some of our long distance clients.

On 20MHz I can normally get around 50 - 60 Mbps to my clients CPE totaling around 70-80 on the sector. But I suspect a large portion of that is because of how far away my clients are from my towers.

25ms latency is normal for SRS. I would recommend staying in SRS if you can. Normal WiFi is hard to diagnose/fix.

Are your Chains 1/2 balanced on the CPE and similarly balanced on the A5? (Like within 6 dB of each other.)


#5

It depends if you’re in Auto mode or Fixed frame mode. One of my radios is actually at 1170 for TX. Here is what you should see in Auto mode with stats like this:

59%20PM

If you’re in fixed frame mode, you should be seeing around 250/250 aggregate.

As for your A5/C5 link in SRS mode, if you’re at full modulation rate and 80MHz channel, you should see around 350/350 TCP.

40MHz is around 175/175 TCP
20MHz is around 80/80 TCP

TLDR: Run your B24 in Auto mode and try your TCP testing again. Please let me know what you see.


#6

@DustinS, I am bookmarking that and will be referring to it in the future…


#7

We’re using 40Mhz, but not seeing any difference in speeds between 20Mhz and 40Mhz channels, even though the PHY doubles. Points to the bottleneck being upstream of them. Looking at the B24 or an interaction with it (Netonix or Mikrotik switches to B24) being the issue.

Though, what confuses me is even if we have a bottleneck at the B24 [that feeds the A5c] of 200Mbps, then we shouldn’t be hitting a wall on the CPEs of ~50Mbps. The CPEs all have -39 to -45dBm on 40Mhz channels, with SNRs of 30-32 ish.

We are in Auto Traffic Split and TDMA mode on the B24 and still hitting a wall at right around 200Mbps total TCP throughput.


#8

Very interesting. Then I would say then you either have an issue with the B24 link itself or something upstream from there. I would be curious what the results would be if you took out the Netonix altogether and used a standard PoE on the B24 (if that’s something easy to do).


#9

Hi,

What method are you using when doing the tests? Number of concurrent streams when doing TCP throughput test is important. Se below example from A5c/C5 using iperf. Try crank up number of streams to like 10 and see what results you get.

SRS
1 stream 128Kb window size
image
2 streams 128Kb window size
image

WiFi Interop
1 stream 128Kb
image
2 streams 128Kb
image


#10

I haven’t been able to iperf across the link yet, we’re using Mikrotik’s built in Bandwidth Test. I’ve been testing with 10 stream and 20 streams.

With 10 streams we see about 170Mbps across the B24 and with 20 stream we see about 200-220Mbps.


#11

Mikrotik hex can’t can do More then 220MBit TCP with Bandwidth Test look at CPU usage, Bandwidth Test is not able to use Multicore CPUs


#12

@Norman1, I was strongly suspicious of this, just didn’t have a Hex to test with.