Community discussions

MikroTik App
 
gdanov
Member Candidate
Member Candidate
Topic Author
Posts: 150
Joined: Thu Jan 17, 2019 1:10 pm

Improving inhouse speed

Sun Jan 03, 2021 5:01 pm

TL;DR: I'm trying to connect two rooms in my apartment wirelessly and so far the best I get is ~250Mbps in the 5ghz band.

Where I live the walls & floors are very thin and the 2.4 spectrum is absolutely packed. The 5ghz is not that crowded yet and still I get extremely disappointing speed when connecting two MT routers over wifi.

I have spent many hours fine tuning the settings and from very slow & unreliable connection I've come to the point where the two routers can hold reliably connection over 80Mhz channel at frequency where I can't hear other stations. I use only two chains on each router (I have hap ac2 and rbm33g with the 3-chain card) and the registrations table reported speed is 650Mbps-80MHz/2S/SGI, however once I stress the connection it typically drops and does not go above 400mbps while the transfer speed itself ranges between 200 and 300mbps. The signal is ~50dBm (I have reduced the power on rbm33g, but that does not make any difference) while the SNR is ~50dB.

Decreasing the AMSDU Limit helped a lot with the stability. Currently at 2048. I'm aware it decreases the bandwidth.

The rbm33g is currently using high-ish gain omni dipole antennas from Alpha (guess ~5db) and from all the reading my understanding is that this may be part of the problem — they pick up too much noise probably.

I'm wondering if changing to lower gain antennas would improve the situation or if using directional antenna like this one https://www.amazon.com/Alfa-APA-M25-dir ... %3D&sr=8-4 will be improvement.

For number of reasons cable or powerline are not practical in my case.
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2978
Joined: Mon Apr 08, 2019 1:16 am

Re: Improving inhouse speed

Sun Jan 03, 2021 6:14 pm

Not easy to suggest settings if not all physical elements are known.

So just some thoughts and experience.

- Directional antenna in 5 GHz to connect between 2 indoor rooms was not an improvement in my house. wAP ac was better than Omnitik ac and that was even better than SXTsq ac.
(My guess: Lower gain antenna sends out more signal power than directional antenna, because the directional transmission is limited to the legal EIRP in the strongest direction. And the connection (multi-)path is unknown. We think it goes straight through the wall in the direction pointed at, but this is not confirmed in the experiments.)
- You did say no one is seen transmitting in the freq used. But you did not say what frequency. Depending on the region, the higher freq is 7 dBm stronger than the lower freq. They are sometimes labeled as outdoor versus indoor. Your installation should be "any" as you are allowed to use all those frequencies indoor.
- Your signal is strong enough (50 dB SNR). It can be too strong and cause deformation because of that.
- 650Mbps-80MHz is MCS7 (64QAM). This should, in non-disturbed condition, be MCS9 (256QAM). So the PHY is stepping down in modulation (because of non-successful transmissions).
- Under load it goes down till 390Mbps. That is MCS4 (16 QAM). The much lower coarse modulation of 16 QAM indicates quite some distortion!
- Your data transmission is not only related to the PHY speed , but also to the TX/RX CCQ % of successful transmissions. (An average of the TX frames / TX HW Frames ratio) (All to be found in the "Registration" tab of wireless. The most important checkpoint.)

What to experiment? Don't just change them all at once. Actually most are even unknown if you should "increase" or "decrease" a value.
(4 parameters with 3 or 2 possible values give 36 combinations to test !)
- different TX power settings. (Avoiding distortion and interference by reflection. Be aware of any metal surface. Don't operate in a corner.)
-"HW retries". Standard is 7. Setting it to 15 will keep the PHY longer in the higher modulation but with more retransmissions. Setting it to 3 will make the PHY lower the modulation faster, avoiding those needless retransmissions.
- Set "adaptive noise immunity" to reduce the sensitivity of the receiver).
-Goto 40MHz. The 400Mbps max you can get there, could be with a better CCQ than the one you get with 80 MHz. Retransmissions are waste of airtime. Your usable PHY is = "interface rate" * "CCQ"
-Lowering A-MSDU is a good action if there are many retransmissions. Smaller A-MSDU has greater chance to pass. A-MPDU will aggregate multiple A-MSDU for performance, but retransmits only the failed A-MSDU (not the whole A-MPDU).
- Getting transfer speed that is between 40% and 75% of the usable PHY is normal, and not a "extremely disappointing speed". There are some tweaks to improve this (like raising the "basic rate"), but that comes later.
 
gdanov
Member Candidate
Member Candidate
Topic Author
Posts: 150
Joined: Thu Jan 17, 2019 1:10 pm

Re: Improving inhouse speed

Sun Jan 03, 2021 7:12 pm

Thanks for the reply. I know it's hard problem. Some details on your feedback:
* lowering the channel width helps a bit with the stability (no disconnects, less speed fluctuation) but still disappointing, only 100mbps and CCQ is same as the wide channel
* RX CCQ fluctuates between 40 and 70% — pretty bad I guess
* even if I improve the SNR by shielding the antennas with bug metal cooking pan I don't see major improvement. But less ping and speed variance, but that's all
* yes, I've read about your theory about lower gain antennas, therefore I'm ready to buy the smallest ones I find
* I decreased the power on the rbm33g side to ~60dbm but no change
* yes, the router decreases the transmitting rate due to bad quality. Therefore I played with all the variables you mention and had success both with making the line fail-fast as well as increasing the hw retries to 15 which seems to work best in my case
* adaptive noise immunity is on
* A-MSDU — I have went the whole spectrum from 256 to the max. Shockingly, with the cooking pan shield I can use max value but still don't go above 300mbps

I have been monitoring my environtment — I don't have microwave oven, no correlation with any of the appliances I have. The only possible problem might be my 3 monitors in the same room as the station router

RE frequencies — I've tried everything from 5200 (crowded) to 5500 and 5560 and 5600. 5500 is supposedly empty, but ping times are ~100ms so something's terribly wrong while 5560 and 5600 are OK

I'm experienced with debugging complex systems so I do my best to be methodical and don't change too much at one go, that's excellent and mandatory advice :)
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2978
Joined: Mon Apr 08, 2019 1:16 am

Re: Improving inhouse speed

Sun Jan 03, 2021 11:59 pm

OK. It is difficult to make major progress if the conditions are not ideal.

We are almost at the end of possibilities.

Just some more ....
- surprised by the 5500MHz long latency. Have you checked with "Freq usage" and "Snooper" , to analyse the existing environment? (Scan only sees AP beacons)
- long latency and many retransmits point to interference. HW protection could help here. (CTS self or RTS/CTS).
- 5600 MHz is in the weather radar freq. Any modification will take 10 minutes for the 5 GHz to come active.
- Microwave oven operates at 2.4 GHz
- CCQ below 85% feels cumbersome to me.
- Performance loss by smaller A-MSDU should be compensated by A-MPDU. Normally A-MPDU is active on priority 0 only. If there are no added rules in the Firewall to set the priority everything is priority 0.
- However these A-MSDU and A-MPDU (A-MPDU only on/off, not a number as with other brands) look very "n"-isch. They use values for 802.11n. The 802.11ac limits are not seen here. And the beacon only reports very small A-MSDU and A-MPDU 'max' sizes. (They are legal 802.11ac values, but the smallest possible.) I don't know if any of these settings are properly used. MKT also gives no counters on MSDU or MPDU.
- 802.11ac is also using VHT settings, not HT settings. In HT settings you can avoid the highest and lowest MCS rates (lowest not recommended here yet), what cannot be done anymore in 802.11ac.
- You can only reduce and avoid the highest MCS rates in VHT. There is one line per chain. You can set MCS 0-7/0-8/0-9 per chain. VHT Basic rates is usually MCS 0-7 only on one chain. (HT uses MCS15 and MCS23 for 2x2 and 3x3 MCS7. HT MCS8 is 2x2 MCS0)
- As you cannot hold the ac rates VHT MCS8 and VHT MCS9 , disabling them , or even just using 802.11n could be an experiment.
- I improved the CCQ in some cases by using the "nv2" protocol instead of the 802.11ac. (You lose the possibility to use the radio for other virtual WLAN functions). But it was from 85%-> 98%. The nv2 is using self-correcting encoding. You lose the view on the number of frames versus HW frames. But whenever CCQ drops often below 90% in PtP and PtMP I switch to nv2.
- I don't need to tell you to only use WPA2/aes-ccm encryption with 802.11. Just checking.

I think we have almost all parameters now.
 
gdanov
Member Candidate
Member Candidate
Topic Author
Posts: 150
Joined: Thu Jan 17, 2019 1:10 pm

Re: Improving inhouse speed

Mon Jan 04, 2021 3:08 pm

I rebuilt my station router to use the steel box I have for this routerboard and stability increased. Then I put together second identical set-up (minus the nice box) to act as AP and right now stability is very good, ping jitter is minimal, but top speed is 400MBps which is still not what I expected. PHY rate is very stable under load. Interesting is that the average throughput fluctuates from run to run (iperf3) from 300 to 380-390 without me changing anything. This is huge variation. See the screenshots bellow.

The wireless snooper at one point showed 250 lines on the 5500 frequency, 80% of them bellow 80db, this sounds crazy. I'm not sure how I could be picking up this. The only explanation is that I live on top floor in densely populated central area...but still...unless there's office building very close by (how close should that be?) that I'm not aware of I can't see how this could be happening. 5600 seems to have radar as the AP refuses to listen on it unless in superchannel mode. I guess in EU we don't have other free channels in the 5500+ range?

interesting is also the iperf3 reports some retries — 100-700 for two-minute test. I was thinking that frame-keepalive can interfere with tcp's flow control but it does not look like this is the case right now.

The state at the moment is much better than the starting one and to big degree thanks to the posts I read from you in different other threads. Still, I would've expected higher and more reliable throughput.

tested NV2 for a moment. Ping was tad slower but no jitter. Throughput was lower (yes, I know it's time division, etc). Right now AC is much better.

cts to self does not seem to make notable difference. I guess I can leave it on.

here is my AP config. Station is identical, it seems important that the two match, esp. re the advanced tab values
/interface wireless

set [ find default-name=wlan1 ] adaptive-noise-immunity=ap-and-client-mode antenna-gain=3 band=5ghz-onlyac channel-width=\

    20/40/80mhz-eCee default-forwarding=no disabled=no disconnect-timeout=1s distance=indoors frequency=5520 keepalive-frames=\

    disabled max-station-count=150 mode=ap-bridge multicast-helper=full name=AP radio-name=RBM33G#2 rx-chains=0,1,2 ssid=\

    myssid tx-chains=0,1,2 wireless-protocol=802.11

/interface wireless security-profiles

set [ find default=yes ] authentication-types=wpa2-psk eap-methods="" mode=dynamic-keys supplicant-identity=MikroTik

https://photos.app.goo.gl/ncfmStQbNoZwJjWb9
https://photos.app.goo.gl/c9yvpMJHbjeRuibx7
https://photos.app.goo.gl/dUaj16MNLLtYfDzN8
 
quackyo
Member Candidate
Member Candidate
Posts: 173
Joined: Mon Nov 16, 2015 10:14 am

Re: Improving inhouse speed

Mon Jan 04, 2021 4:37 pm

You say you reduce TX power on on of the sides, but did you reduce it on both sides? If not, you should test that too.
Other than that, I am not surprised that 5Ghz inhouse through a wall is not 100% stable @ 80MHz.
 
gdanov
Member Candidate
Member Candidate
Topic Author
Posts: 150
Joined: Thu Jan 17, 2019 1:10 pm

Re: Improving inhouse speed

Mon Jan 04, 2021 4:49 pm

yes, reduced on both sides, then set to nominal. currently rssi is 50db.

if you read my last post you'll see that it's currently very stable regardless if I use hap ac2 or rbm33g + R11e-5HacT as AP. Station is always rbm33g + R11e-5HacT
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2978
Joined: Mon Apr 08, 2019 1:16 am

Re: Improving inhouse speed

Mon Jan 04, 2021 6:04 pm

Good job !
Interesting is that the average throughput fluctuates from run to run (iperf3) from 300 to 380-390 without me changing anything. This is huge variation. See the screenshots bellow.
Your RF spectrum around might be fluctuating. (The other radio's around). That 300-390 Mbps is not a bad number. The higher the PHY the higher impact the overhead has. On stable high PHY some steps must be taken to reduce the impact of the overhead. One is rising the basic rate on all radio's. (from 6Mbps to 18 Mbps or even 24 Mbps).
The wireless snooper at one point showed 250 lines on the 5500 frequency, 80% of them bellow 80db, this sounds crazy. I'm not sure how I could be picking up this. The only explanation is that I live on top floor in densely populated central area...but still...unless there's office building very close by (how close should that be?) that I'm not aware of I can't see how this could be happening.
Unfortunately that is your reality. If you allow "Receive errors" in Snooper, you will even see more! You use everything on 5500-5520-5540-5580 with your setting. Co-channel interference happens with every sender that is 6 dB above the noise floor. So you might even wait for someone sending at -96 dBm. Fortunately 802.11ac does a better job than 802.11n for wide channels, by using only some of the 4 channels if needed. (With speed variation ...)
Check your antenna pattern (sending=receiving). Even a 3 dBi low gain antenna has already some donut shape. (there is lower signal in the length axis of the antenna). This is an expliciete donut shape for higher gain antenna. If you have external antenna, set them all upright, not under an angle like in the commercial pictures.
5600 seems to have radar as the AP refuses to listen on it unless in superchannel mode. I guess in EU we don't have other free channels in the 5500+ range?
If you have radar signal detected this will be clearly noted in your log. (Only the AP does radar detect handling). For wifi visibility extend the log with (/system logging add topics=wireless action=memory). All channels above 5500 in EU (Etsi regulatory region) are still DFS channels. The weather channels (5600-5620-5640) take 10 full minutes before they activate after every change of the interface or wireless setting. 10 minutes is extremely long ... if you have to wait for it. The LOG again is your guidance. It will tell you "searching for radar" and "starting network" (not the exact wording ).
There is no need to assume you have radar, you can see it in the log. Actually there are also many "false radar detects".
interesting is also the iperf3 reports some retries — 100-700 for two-minute test. I was thinking that frame-keepalive can interfere with tcp's flow control but it does not look like this is the case right now.
Normally wifi does not drop any packet. It retries until disconnect. TCP congestion avoidance can play a big role in here. Depending on your client you can experiment with the algorithms. https://en.wikipedia.org/wiki/TCP_congestion_control . MKT does not allow us to specify the algorithm, as Linux client does, and Windows 10 has changed to CUBIC I believe. MKT is not sending TCP data itself, tests should be with sending/receiving clients.
The state at the moment is much better than the starting one and to big degree thanks to the posts I read from you in different other threads. Still, I would've expected higher and more reliable throughput.

tested NV2 for a moment. Ping was tad slower but no jitter. Throughput was lower (yes, I know it's time division, etc). Right now AC is much better.
That's why I start with 802.11ac and only go to nv2 with poor CCQ.

cts to self does not seem to make notable difference. I guess I can leave it on.
I have to reread the 802.11ac survival guide. 802.11ac may be using RTS/CTS to handle separate channels in wide channels. https://www.oreilly.com/library/view/80 ... index.html (free). See https://www.oreilly.com/library/view/80 ... /ch03.html for RTS/CTS. The other chapters are interesting too.
here is my AP config. Station is identical, it seems important that the two match, esp. re the advanced tab values
Well I don't like the 6.48 as RouterOS version. Too many reported problems !!! Use the LT (Long Term).

You can add screenshots and photo's here in "Attachments" of a post , they can even be placed "in line".
 
gdanov
Member Candidate
Member Candidate
Topic Author
Posts: 150
Joined: Thu Jan 17, 2019 1:10 pm

Re: Improving inhouse speed

Mon Jan 04, 2021 6:35 pm

thanks again!

my antennas are 3x dipole omni. On one box I have them in line, mounted on the back of mikrotik box, several cm apart. On the other they are in triangle pattern, 10cm apart. I keep them vertical ATM. Setting gain between 3 to 7 gives me the same performance.

downgraded, run quick test — no change.

I got the snooper numbers with receive errors on...There are couple of office buildings around, but I hoped they are far enough. I guess the alfa networks antennas I use are actually good (specd 5db gain).

I'm running only macs and linux at home. My point was that by setting frame life to some low value I'll give chance to TCP flow control to kick in. but with the 70% ccq I have under load I guess I shouldn't bother.

You mention the datarates several times. How am I supposed to raise the VHT rates (I'll stick to AC-only) ?
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2978
Joined: Mon Apr 08, 2019 1:16 am

Re: Improving inhouse speed

Mon Jan 04, 2021 9:43 pm

my antennas are 3x dipole omni. On one box I have them in line, mounted on the back of mikrotik box, several cm apart. On the other they are in triangle pattern, 10cm apart. I keep them vertical ATM. Setting gain between 3 to 7 gives me the same performance.
That's OK then. You know that by setting antenna gain in RouterOS you only change the max TX power of the radio = (ETSI max limit - antenna gain set)

Just one step further. In a crowded environment with all the interfering neighbors on one side, I reduced once the interference by having the antenna parallel but under a 30° angle from vertical, away from that side. (Signal and reception high above their head. They used long range outdoor units (26dBm TX power +3dB (for 2 antenna) +8 dBi antenna gain = 37 dBm in ETSI region)
downgraded, run quick test — no change.
OK
I got the snooper numbers with receive errors on...There are couple of office buildings around, but I hoped they are far enough. I guess the alfa networks antennas I use are actually good (specd 5db gain).
Is OK if they are for 5 GHz.
I'm running only macs and linux at home. My point was that by setting frame life to some low value I'll give chance to TCP flow control to kick in. but with the 70% ccq I have under load I guess I shouldn't bother.
Not sure I understand where you set "frame life" , and how this would change the TCP flow control. You mean "frame lifetime" in MKT wireless ?
This is a very special parameter! Wifi retransmits a packet by retries (3 times # of HW retries, then steps down one MCS level) repeat until MCS0, if that double fails, then retries during "Disconnect timeout" , if it still fails then disconnects and drops the packets for that connection. Wifi normally does not drop packets. TCP flow control is done over negotiations (window, buffer, RTD, missing ACK). See https://wiki.mikrotik.com/wiki/Manual:Wireless_FAQ for the MKT implementations (other brands use other algorithms for retransmits and rate adjustment, sometimes you have an option to choose)


You mention the datarates several times. How am I supposed to raise the VHT rates (I'll stick to AC-only) ?
I don't know why it is hidden in the GUI. If you set it as "configured" under data rates, then it appears.

Default values:
Klembord-1.jpg
To be tuned like this for less overhead:
Klembord-2.jpg
To estimate the impact: use http://www.arubanetworks.com/assets/vrd ... RD_V2.xlsx (error in right table, Data rate A-MPDU)
You do not have the required permissions to view the files attached to this post.

Who is online

Users browsing this forum: Amazon [Bot], Cloudtechiq [Bot] and 30 guests