Community discussions

MikroTik App
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

802.11ac Really Poor Performance?

Fri Feb 25, 2022 11:16 pm

I am testing two setups:

1) RB922UAGS-5HPacD.
2) RMB33G with a R11e-5HacD radio module.

I have set them up on 80MHz wide channels with 5GHz-only-AC band setting. When I run the Mikrotik bandwidth test using UDP packets the best throughput I see is around 100 to 145Mbps Transmit and the same receive. At best this is around 290Mbps total throughput. AND this is NO WHERE NEAR the 867Mbps throughput these radios are supposed to be capable of.

Test is not over air - it is through coaxial cables and attenuaters (70dB of attenuation). Signal Strength around -50dB - I have tried using different protocols and settings but nothing seems to improve the wireless data rate.

When I had tested the RBM33G with the R11e-5HacD card a few years back the 802.11 wireless protocol setting was giving me around 420 to 480Mbps throughput over the same attenuator setup I am using now.

Couple that with the fact that the new RB922UAGS-5HPacD has an additional problem in that - no matter what I set the Tx Power to (6dB up to 30dB) the signal strength basically does not change. It is like the Tx Power is stuck at a moderate setting and acts like it is no where near the 30dB max output power advertised for this board.

This is really substandard performance and I am evaluating these radios for inclusion in our small companies product line - this this would equate to the sale of over 500 of these in the next year or so.

If I cannot get an answer or any support we will have no choice but to drop these products as the performance is no where near what manufacturer claims.

Question: Is anyone else seeing this? Is this standard for this product?

If you need a settings file I will be happy to provide it.

Hopefully someone will respond????

Wireless Settings are below:
[admin@922 - EmBedded AC-AP] /interface wireless> print detail
Flags: X - disabled, R - running
0 R name="wlan1" mtu=1500 l2mtu=1600 mac-address=2C:C8:1B:EF:F6:D2 arp=enabled
interface-type=Atheros AR9888 mode=ap-bridge ssid="USELESS1" frequency=5765
band=5ghz-onlyac channel-width=20/40/80mhz-Ceee secondary-channel=""
scan-list=5175-5255,5740-5850 wireless-protocol=nv2 vlan-mode=no-tag
vlan-id=1 wds-mode=dynamic wds-default-bridge=wds-bridge wds-ignore-ssid=no
bridge-mode=enabled default-authentication=yes default-forwarding=no
default-ap-tx-limit=0 default-client-tx-limit=0 hide-ssid=no
security-profile=Radio1 compression=no
 
mducharme
Trainer
Trainer
Posts: 1777
Joined: Tue Jul 19, 2016 6:45 pm
Location: Vancouver, BC, Canada

Re: 802.11ac Really Poor Performance?

Fri Feb 25, 2022 11:55 pm

If you are running the bandwidth test from the same device that is doing the wireless, the bandwidth test will give you a reduced result because it uses so much CPU on the device. This will reduce the wireless capacity below what it could handle for actual traffic.
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Sat Feb 26, 2022 12:22 am

WOW - has this always been the way the Bandwidth test tool on the Routerboard works? If you run it on the Access Point or a Station it gives reduced results?

So what setup do you suggest to run bandwidth test to get an accurate result. Iperf - from PC to PC over the wireless?

If this is the case other than an external bandwidth test tool like iperf - what kind of setup do you suggest using the Mikrotik bandwidth test tool that will provide accurate results?

Thanks for the reply!
 
mducharme
Trainer
Trainer
Posts: 1777
Joined: Tue Jul 19, 2016 6:45 pm
Location: Vancouver, BC, Canada

Re: 802.11ac Really Poor Performance?

Sat Feb 26, 2022 12:29 am

WOW - has this always been the way the Bandwidth test tool on the Routerboard works? If you run it on the Access Point or a Station it gives reduced results?
So what setup do you suggest to run bandwidth test to get an accurate result. Iperf - from PC to PC over the wireless?
If this is the case other than an external bandwidth test tool like iperf - what kind of setup do you suggest using the Mikrotik bandwidth test tool that will provide accurate results?
Yes it has always been like this. iperf is fine. The way of doing it with MikroTik bandwidth test on a routerboard is to have a separate routerboard on both sides that is doing nothing but btest over the link, with a more powerful CPU (ex. ARM or ARM64). In many cases this is less convenient than using iperf as you may not have these extra routerboards to use for this, or they may not be powerful enough to generate the traffic needed to saturate the link.
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Sat Feb 26, 2022 12:43 am

I have extra routerboards so I can setup and run the test the way you suggest. I will give that a try.

I have nothing but the Access Point connected to my PC and I turned off everything in the Ethernet Settings but TCP/IP v4. So windows traffic is little to none!

Made sure the Ethernet connection was negotiated at 1Gbps and ran the test over the wireless to the station using the btest tool. I did not see any better results for this setup.

Actually the results were almost identical to the onboard tests.
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2983
Joined: Mon Apr 08, 2019 1:16 am

Re: 802.11ac Really Poor Performance?

Sat Feb 26, 2022 2:42 am

AND this is NO WHERE NEAR the 867Mbps throughput these radios are supposed to be capable of.
867Mbps is the interface rate , the PHY, the rate at which the radio is transmitting. This is far from the possible "data rate", the payload rate, the effective throughput of user data. 802.11 has a lot of overhead to handle the CSMA/CA , the timing of the data packets. This overhead becomes hughe at the higher interface rates, because the overhead remains at the same rate.

viewtopic.php?t=165698#p912622 ; viewtopic.php?t=165698#p817145
But in real life you never reach that island-performance because .... viewtopic.php?t=183101#p915424
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Mon Feb 28, 2022 6:52 pm

AND this is NO WHERE NEAR the 867Mbps throughput these radios are supposed to be capable of.
867Mbps is the interface rate , the PHY, the rate at which the radio is transmitting. This is far from the possible "data rate", the payload rate, the effective throughput of user data. 802.11 has a lot of overhead to handle the CSMA/CA , the timing of the data packets. This overhead becomes hughe at the higher interface rates, because the overhead remains at the same rate.

viewtopic.php?t=165698#p912622 ; viewtopic.php?t=165698#p817145
But in real life you never reach that island-performance because .... viewtopic.php?t=183101#p915424
WOW - I tried to read all three of those posts and it was way over my head - your claim of 366Mbps to 600Mbps I would be totally happy with for bandwidth across the 802.11ac link.

When I preformed the same tests three years ago on a RBM33G with an R11e-5HacD radio module I was seeing 400 - 480Mbps over the wireless connection which I was very happy with it - and our company approved the product for sale.

I am now evaluating the RB922UAGS-5HPacD and I am seeing between 200 and 300Mbps - when I re-ran my tests on the RBM33G set up with the R11e-5HacD radio - I am seeing the EXACT same results.

I was told it is processor utilization issue - Running Iperf, and routerboards on either side of the 922 boards (as suggested above) leaves me seeing the same results, and running btest tool from my PC to the remote radio gives the EXACT same results. So I am left to conclude the the processor utilization issue - is not the correct cause of the problem.

Your posts were quite complicated and did not provide a clear direction to go in terms of solving my problem. About the most I could get out of the posts is that I should set the AMSDU limit higher and check off more AMPDU priorities on the HT tab in the Interface <wlan1> window? Or am I totally missing the point about these settings.

In writing this I tried to adjust the AMSDU limit and threshold - and I am unable to enter a setting higher than 8192 - checking off additional AMPDU priorities does not improve the bandwidth test results at all. So the net result of your posts and information is that it has no effect on my bandwidth test results.

Do you or anyone have any suggestions on what settings need to be programmed into Router OS/Winbox to increase the bandwidth above 200 - 300Mbps? Or is that it? That is what the RB922UAGS-5HPacD run over air - and there is nothing I can change in the wireless configuration to improve the test results?
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2983
Joined: Mon Apr 08, 2019 1:16 am

Re: 802.11ac Really Poor Performance?

Mon Feb 28, 2022 8:51 pm

Well that was the indepth analysis of what is (assumed to be) going on. It is partly reverse engineering (the 8 MPDU in one A-MPDU is just using numbers to have the measured values)
Some of it is defined. Like the supported max MDPU and A-MPDU are broadcast by the AP 10 times per second. And Mikrotik uses the low values of the 802.11 standard.
There is nothing we can do about this in RouterOS. Well I have no idea how to set this in MT. There is also no feedback on aggregation used.

What can be learned from this are a confirmation of the data rate numbers measured (Mikrotik)
400Mbps-40MHz/2S/SGI can deliver 260 Mbps as data rate. This will be reduced by the environment. (retransmits, busy airtime, noise, destructive interference), and due to waiting for a clear transmit channel.
800Mbps-80MHz/3S/SGI goes up to 360 Mbps. Again with reductions as above
1300Mbps-80MHz/3S/SGI goes even higher, but I have no measurements, as I have no client devices that support 3 streams.

Improvements that can be made are small. Higher basic rates if condtions are very good is one.
Changing software (wifiwave2, OpenWRT, other vendor ...) will use bigger MPDU and A-MPDU, and this delivers more data for the same interface rate.

Possible values for MPDU and A-MPDU are in the 802.1 standards.
Packets are combined into MSDU, -> aggregated into A-MSDU what is the payload for the MPDU, -> that agregates into A-MPDU.
What the AMSDU parameter setting does in Mikrotik is not documented. The MPDU used is 3895. Corresponding A-MSDU is 3839 (for 802.11n next value is 7935, 802.11ac also has 11462)
But how could 8192 or a higher than 3839bytes AMSDU size fit in that MPDU ? Lower value will probably reduce the size transmitted. AMSDU=8192 does not alter the MPDU size used.

Using WMM with A-MPDU disabled will give small packet delays (no buffering) but much lower throughput. Enabling and using WMM on Mikrotik is not that simple.

One extra remark: "nv2" is NOT a 802.11 protocol ! The overhead is totally different (much smaller, there is no contention window, no clear channel assesment and competition to get access to transmission). "nv2" can work with much smaller transmissions without falling to a very low data rate like 802.11 does. (https://mum.mikrotik.com/presentations/IN12/soumil.pdf)
.
Klembord-2.jpg
You do not have the required permissions to view the files attached to this post.
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Mon Feb 28, 2022 10:36 pm

WOW you are really well educated and informed on this topic! I really appreciate you taking the time to respond to my post.

So like you said in your last post. There is really no way to change any settings in Winbox to increase the speed of the wireless link to the advertised MCS07 through MCS09 speeds?

Thanks again!
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Mon Feb 28, 2022 10:53 pm

WOW - has this always been the way the Bandwidth test tool on the Routerboard works? If you run it on the Access Point or a Station it gives reduced results?
So what setup do you suggest to run bandwidth test to get an accurate result. Iperf - from PC to PC over the wireless?
If this is the case other than an external bandwidth test tool like iperf - what kind of setup do you suggest using the Mikrotik bandwidth test tool that will provide accurate results?
Yes it has always been like this. iperf is fine. The way of doing it with MikroTik bandwidth test on a routerboard is to have a separate routerboard on both sides that is doing nothing but btest over the link, with a more powerful CPU (ex. ARM or ARM64). In many cases this is less convenient than using iperf as you may not have these extra routerboards to use for this, or they may not be powerful enough to generate the traffic needed to saturate the link.
I just got done setting this up and running your suggested setup. Ran an old RB435G on an Ethernet connection to the RB922 - then over air to the remote RB922 and it is connected over Ethernet to another RB435G - checked and all 4 Ethernet ports were connected at 1Gbps - Ran a bandwidth test from the first RB435G to the RB922 - and it ran at right around 1Gbps - sometimes very close to 500Mbps in each direction. Did the same from the remote RB922 across the remote Ethernet link to the last RB435G in the chain. Results were identical. This confirms that the Bandwidth Test tool is accurate and giving correct bandwidth for a 1Gbps link, also confirms that the RB435G is up to the task of generating enough traffic to saturate the link.

I then ran the Bandwidth test tool to the remote RB435G - Tx/Rx Total Average was 108.6Mbps/110.7Mbps

I then ran the Bandwidth test tool on the first RB922 to the remote RB922 across the 802.11ac wireless link - this test which should be reduced due to processor utilization yielded the results - Tx/Rx 118.0Mbps/113.4Mbps.

So results were basically the same, and at this point I cannot contribute the low bandwidth over the wireless connection to my own user error, or over utilized processor.

Is there a resource anywhere that has suggested Winbox wireless settings that provide better bandwidth then the present settings I have?

BTW - thanks for the help - I do appreciate it, if anything we did prove that the bandwidth test tool is fairly accurate!
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2983
Joined: Mon Apr 08, 2019 1:16 am

Re: 802.11ac Really Poor Performance?

Mon Feb 28, 2022 11:47 pm

When I run the Mikrotik bandwidth test using UDP packets the best throughput I see is around 100 to 145Mbps Transmit and the same receive. At best this is around 290Mbps total throughput.
"Total throughput" !

1. Wifi is always unidirectional. Simultaneous send/receive will be somewhere halve the data rate for send or receive. The 866Mbps interface rate is not like the 1Gbps of the ethernet cable. Ethernet these days over UTP cable or fiber is always bidirectional, with minimal overhead. (It used to be unidirectional or half-duplex on the coax cabling, and the early hubs and switches for utp cables. 100Mbps still negotiates this.)

2. Never deployed "nv2" above "dual stream 40MHz" (around 260 Mbps). Forgot performance numbers for 80MHz wide channel. Tuning should be under "nv2". Frame size is expressed in milliseconds here (TDMA period size). The frame size grows with the speed !? Default 2ms allows for 80000bytes for 400Mbps, and 173000bytes for 866Mbps, if nv2 can handle this. (!)

Hint: disable WDS mode.
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Tue Mar 01, 2022 12:26 am

When I run the Mikrotik bandwidth test using UDP packets the best throughput I see is around 100 to 145Mbps Transmit and the same receive. At best this is around 290Mbps total throughput.
"Total throughput" !

1. Wifi is always unidirectional. Simultaneous send/receive will be somewhere halve the data rate for send or receive. The 866Mbps interface rate is not like the 1Gbps of the ethernet cable. Ethernet these days over UTP cable or fiber is always bidirectional, with minimal overhead. (It used to be unidirectional or half-duplex on the coax cabling, and the early hubs and switches for utp cables. 100Mbps still negotiates this.)

2. Never deployed "nv2" above "dual stream 40MHz" (around 260 Mbps). Forgot performance numbers for 80MHz wide channel. Tuning should be under "nv2". Frame size is expressed in milliseconds here (TDMA period size). The frame size grows with the speed !? Default 2ms allows for 80000bytes for 400Mbps, and 173000bytes for 866Mbps, if nv2 can handle this. (!)

Hint: disable WDS mode.
WOW your as old as I am - that Coax cable reference kind of gives you away - Remember Token Ring? :D

So when you ran UDP bandwidth test you are seeing the same results that I am - well at least that lets me know I'm not headed off into some unknown direction.

Thing that I find confusing is that just three years ago with the RBM33G main board and R11e-5HacD radio card - the MT bandwidth test tool was giving me 400 - 480Mbps across the wireless link - now it is not even close - back when I did those tests NV2 gave terrible results and 802.11 wireless protocol delivered by far and away superior results - here we are trying to approve the 922 board and the tool and wireless connections no longer produce even close the the same results - regardless of what protocol I use.

So far with this round of testing the best throughput I have been able to achieve is using the NV2 protocol - 802.11 for some reason no longer produces the results it did a few years ago!

Well I do appreciate you letting me know that you saw the same bandwidth that I saw testing the MT products.

Thanks!
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2983
Joined: Mon Apr 08, 2019 1:16 am

Re: 802.11ac Really Poor Performance?

Tue Mar 01, 2022 1:22 am

Token ring, sure, Type 1 cabling, logical ring physical star :-). Async and sync modems, even SLIP (before PPP).

I'm aware of performance degradation for 802.11 since 6.45.6 for some newer ROS versions. ROS 7.2RC4 seems to be back on track, from what I read here indirectly, not tested yet.
The AMSDU of 8192 must have been used somewhere in time in some ROS, picking up the 802.11n based 7935 max size, but the devices I have only use a MPDU of 3895.
https://www.oreilly.com/library/view/80 ... /ch03.html
When did they reduce it in ROS? Maybe the units with more memory only, now use bigger buffers?????? That 7935 (or a larger A-MPDU) could go up to 600Mbps for 866Mbps interface rate.
Couple that with the fact that the new RB922UAGS-5HPacD has an additional problem in that - no matter what I set the Tx Power to (6dB up to 30dB) the signal strength basically does not change. It is like the Tx Power is stuck at a moderate setting and acts like it is no where near the 30dB max output power advertised for this board.
These chipsets for ac , do drop with higher MCS rates considerably. MCS0 is 30dBm, but MCS09 is only 22dBm. One must add the antenna gain for the legal EIRP. ( and 3dBm for 2 antenna if 802.11n, not sure about 802.11ac as MT claims to use total in case "ac"). EIRP will be limited to 30dBm (20 dBm in Europe)
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Tue Mar 01, 2022 10:10 pm

I got it! I got the settings that were giving me good bandwidth on the Bandwidth Test tool. The throughput I am seeing is - current Tx=207.4 Mbps/Rx=194.3 Mbps, 10 second average Tx=212.1 Mbps /Rx=194.1 Mbps, Total Average Tx=224.0 Mbps/Rx=172.9 Mbps. This is the bandwidth results I was looking for.

This isn't 866Mbps but it is over 366Mbps and close to 400Mbps which is exactly what I was looking for.

Settings that gave me these results are:
I am using WDS so one radio is set as an "ap bridge" the other set as "station wds"
Band was set to 5GHz-A/N/AC Channel width was set to 20/40/80MHz Ceee or eCee - both gave very similar results
DFS is disabled
WPS mode is push button
Wireless Protocol = 802.11 - note: NV2 gave noticeably lower bandwidth readings.

On data rates tab only the lowest supported and basic a/g rates were selected.
Under the VHT Supported MCS I had a line for MCS 0-9, second line for MCS 0-8 and a third line for MCS0-7, and VHT Basic MCS had a single line for MCS 0-7.
On the HT MCS tab only MCS0 checked for both supported and basic.
WDS was set to dynamic.
I had WPA2-PSK encryption on link.

Thanks to bpwl for all your posts and great information, and mducharme for pointing me toward a more accurate bandwidth test.
 
User avatar
bpwl
Forum Guru
Forum Guru
Posts: 2983
Joined: Mon Apr 08, 2019 1:16 am

Re: 802.11ac Really Poor Performance?

Tue Mar 01, 2022 10:52 pm

OK now let me fine tune this :-). I can whine until it is good. Sorry. First save this as good starting point.

1. MCS setting. MCS0-9 allows rate up to MCS9. MCS0-8 allows MCS up to MCS8 only. I see no reason for this. Both streams/chains should work at the highest MCS rate possible.
(See VHT in https://mcsindex.com/. VHT is used by 802.11ac, HT is used by 802.11n and is different as MCS rates numbers rise with # of streams. 802.11 has MCS set per stream)
The 3th MCS line is for the 3th stream (if there is one) and should also be MCS0-9.
Basic rate is a different story. And MCS0-7 is the usual setting.
Lower setting in the supported rates is only done if the rate selection is flapping for these high rates due to failed transmissions.

2. WPS if not used should be disabled. It enlarges the beacon with information if set.

3. WDS. I hinted not to use WDS. Maybe you really need it for the separate interface and firewall rules. If not needed as separate interface use "AP-bridge" to "station bridge" for the link.. "AP-bridge" - "station bridge" is a transparant 4-address link just as an ethernet link. This is only with RouterOS !!!

4 802.11 protocol. Only enable the protocols you need. "ac" is the fastest for Mikrotik. Remove the "N" if not needed, i doubt if you have slower "A" devices.
HT with only MCS0 checked, that is 802.11n only at 15Mbps for 2S/SGI. See MCSINDEX! Supported: Check them all up to MCS15, basic rate MCS0 till MCS7. (This is almost as VHT then)
HT MCS up to MCS23 for triple stream (3S, 3 chain, 3 antenna)

5 "Band was set to 5GHz-A/N/AC Channel width was set to 20/40/80MHz Ceee or eCee - both gave very similar results"
Ceee has the lowest channel used for the frequency set. eCee has the second lowest channel set to the frequency. 5180/Ceee = 5200/eCee 80 MHz wide channel. Protocol starts by checking the "C" , then the "e", that would be the difference.
 
dbraddon
newbie
Topic Author
Posts: 30
Joined: Thu May 12, 2011 7:51 pm
Location: Boulder Colorado
Contact:

Re: 802.11ac Really Poor Performance?

Wed Mar 02, 2022 12:03 am

OK per your recommendations:

1st test!
1 Set VHT MCS rates to two lines on the supported both MCS 0-9, and basic left it at MCS0-7.

2) Disabled WPS mode

3) Left WDS tab at dynamic and interface at bridge1 (the bridge we configure that we attach the wlan1 and Ether1 ports to.

4) checked off HT MCS supported rates MCS0-15, basic MCS0-7.

5) Left Channel Width at 20/40/80MHz Ceee.

Results: Current Tx=229.0Mbps/Rx=244.5Mbps, 10sec average Tx=242.6Mbps/Rx=215.8Mbps, Total Average Tx=229.8Mbps/Rx=184.8Mbps

2nd Test
Everything the same as 1st test with:

WDS tab set to disabled
Access Point set to ap bridge, and station set to station bridge.

Results: Current Tx=305.5Mbps/Rx=118.8Mbps, 10sec average Tx=285.7Mbps/Rx=133.1Mbps, Total Average Tx=245.4Mbps/Rx=147.8Mbps.

I let each bandwidth test run for 3 minutes

With WDS disabled the Tx rate from the AP to the Station was substantially improved - Rx rate didn't do as well.
With WDS set to dynamic and station set to station wds Tx was not as good, and Rx was slightly better.

Since our companies default settings are WDS set to dynamic and WDS interface set to bridge1, and station set to station wds - customers don't really need to turn off wds and they still get good results. So my suggestion to customers is to leave WDS alone. Actually, I won't even bring up the topic.

WOW - you are a huge help - I cannot thank you enough for your support. If it hadn't been for you I would not have solved this problem! Thank you very much!!

Who is online

Users browsing this forum: Amazon [Bot], ccz117, jhbarrantes, petardo and 9 guests