Community discussions

MikroTik App
 
User avatar
bpwl
Forum Guru
Forum Guru
Topic Author
Posts: 2978
Joined: Mon Apr 08, 2019 1:16 am

How to get rid of "slow wifi". Can it be done ?

Sun May 17, 2020 7:32 pm

The cry is al over the wifi world "help my wifi is slow". And Mikrotik devices don't have the solution built in.

Its starts with the expectations. On the box are the aggregated bandwidths (AC2600 etc). Only later one discovers that actually this is the sum of the maximum interface rates.
Only much later one learns that those maximum interface rates are for a near range communication in free field or in a room with no reflections, diffraction, absorption or interference only.
That last one is hard for people that only look to their own equipment. Their location is free from other transmitters, they are quite certain of this. Once they use their measuring tools they can see that they are limited in bandwidth, air-time and still be disturbed by noise from a lot of radiation they didn't know about.
They eventually land on a world realistic channel bandwidth, MCS encoding, number of spatial streams giving them a usable interface rate. Quite far from that AC2600 number.

So with the expectations set to a more realistic number , we are not done yet with the disappointments. Going from interface rate to the data rate is still a big step down.
The media-access overhead and the management air-time is taking out a very large part of that "interface rate". 60% of the interface rate is the data rate say some papers.
Well in one direction. Less than half in both directions. And if there are further away clients (lower MCS rate), or clients with lesser chains (spatial streams), they will be slow.
They will almost all become that slow in data rate as the slowest client, that get most of the air-time.
And WDS will halve that rate, just as a same-channel-repeater does. And repeater on repeater will be 1/3 of the speed.

And its not all yet. The conditions are not stable. Even more, the AP and clients have not enough information to find the highest achievable rate, even under stable conditions. So they will select a higher rate, until that one fails to deliver, and then step or drop down the rate. To later step up again. Do this in ever changing conditions and it all becomes very unstable. And as with all rate mechanisms the impact on the average rate of the slowest rates is higher than the impact of the higher rates.

And then we have these high interface rates. That makes the whole mechanism of getting an effective data rate very difficult. Where deficiencies at rates up to 56Mbps could be tolerated, the same deficiences become a major problem at 300 ... 433 ...866 ... 1733 Mbps. To have at least some usefull percentage of data rate out of high speed interface rates , extra steps must be taken . Things like A-MSDU and A-MPDU are crucial. But these aggregation techniques loose their value if there are transmission errors. And without them the % data rate is very low.

So even if you set up the wifi environment , while measuring and tuning (viewtopic.php?f=7&t=161170#p793560) you still might end with lower values than expected.
And then we start tuning. Tuning however is like "ordering a smaller pie, in such a way to have a larger share of the pie, and end up eating more pie". The steps we take in tuning is deliberatly lowering the capacity, but getting a larger % of that lower capacity, and ending up with a larger throughput. This makes tuning by copying "very performant configurations" from others almost a certitude to fail. I have seen in this forum people halving the bandwidth because it helped with others, reducing the A-MSDU because it helped with others, changing the lower basic rates but also the lower supported rates because it helped with that other config, etc etc etc ... And it did indeed help with those others, but the combination of these uncorrelated settings gives only one certainty: the capacity is seriously reduced, and there is no correlation to the own situation, so there is no larger share.

Why is it so difficult? Why do the tips&tricks of the old A,B,G times not work? For me because it has become so complex and so dynamic that one can tune it right for one environment, if that environment is steady, and for one type of usage. No one has problems with a radio free spectrum, but it get complex if the requirements of the differnet users varies, and if the load in the wifi-spectrum changes, or if the link AP-client is not free from artifacts. Our only hope is on the algoritmes in the AP and in the client. They don't have much information (signal strength, error rate, ack-timing ...) but could adjust continously. I don't think you could script it in Mikrotik. But maybe some of it could be done.

The high speed interface rate challenges....
Klembord-1.jpg
If no aggregation is used, or if A-MSDU is set low, like 1000, this is the % you get. At 300 Mbps we have only 7.5% left. And what would it be at 866Mbps ?
So aggregation is important,
Klembord-2.jpg
Even at 54 Mbps the impact is clear. And 1000byte packets are not the smallest in real life.
Klembord-3.jpg
But you can overdo it. In 802.11n the A-MPDU size was 64KB, and 128KB for 802.11ac. (I think that's what I saw, but I lost track of that information)
Challenge with Mikrotik I have, is the limited information. By default it is activated for priority 0 only ("best effort"). Not for all the rest.
Is WMM setting priorities? Or should it be done in the Firewall Mangle rules? Never done, not seen elsewhere. I have no idea where to look to see the a-MPDU in action.
If nobody sets priority is everything "0" then, or "void". Should we be sniffing and using Wireshark? Or is there another way? (Indirect by measuring throughput is not the proper way of doing this.)
Klembord-4.jpg
(144 MBPS interface rate) The A_MSDU , more efficient than A-MPDU, fails if there are bit errors in the transmission.
It can go very low. Actually all those number are low, for the user expectation of 85% of the interface rate. And we are here only at 144Mbps!
Without an automatic adjustment for A-MSDU and A-PMDU based on the error rate we are in for some major disappointment.


Klembord-5.jpg
There is quite a difference here between "Adaptive frame aggregation" and fixed frame aggregation.
Certainly if you set the frame aggregation low (A-MSDU=2048) comes near to the bottom line, if A-PMDU does not act.

But I don't know what the RouterOS is doing.
I don't understand the interaction between WMM and the priorities for A-PMDU
Is there an algoritme behind the A-MSDU setting. It's negotiated , but anything else?
I know, aggregation and respons-time/jitter or opposite sides of the tuning. It's the one or the other.
Can WMM differentiate? Via priorities? Otherwise ?


Can we ask Mikrotik to look in those algoritmes. Do they have free access to them. Can they be implemented.
And while I see those studies are closely related to "airtime fainess" we might be very lucky to get this feature as well.
I find airtime-fairness an important feature in a larger non-related group. But with the current RouterOS I do not see how I could simulate such behaviour.
As studies, I mean things like this: https://www.hindawi.com/journals/misy/2015/548109/


The other information came from:
http://ceur-ws.org/Vol-1689/paper7.pdf
http://www.ece.ubc.ca/~vincentw/C/LWcGC06.pdf
You do not have the required permissions to view the files attached to this post.
 
User avatar
bpwl
Forum Guru
Forum Guru
Topic Author
Posts: 2978
Joined: Mon Apr 08, 2019 1:16 am

Re: How to get rid of "slow wifi". Can it be done ?

Sat May 23, 2020 12:32 pm

Ok already some answers in this forum.

-priority marking is not done by WMM itself. DSCP is not used by WMM. So WMM needs the mangle rules to have the priority set. (from: viewtopic.php?f=2&t=125152)

-setting priority via mangle rules can require a long list of rules, and many queues. QoS has to be configured by specifying those priorities. (from: viewtopic.php?f=23&t=73214)

-The relation to A-PMDU is clear in the priorities from the mangle rules, but its implementation is not known. How many A-MSDU/MSDU packets in one A-PMDU? Fixed? Variable? Value per priority ?

The priority setting of WMM looks very similar to the "nv2-queue=4" in the nv2 frame priority. .... https://mum.mikrotik.com/presentations/ ... 536100.pdf
 
User avatar
bpwl
Forum Guru
Forum Guru
Topic Author
Posts: 2978
Joined: Mon Apr 08, 2019 1:16 am

Re: How to get rid of "slow wifi". Can it be done ?

Wed Jun 10, 2020 11:15 am

There are quite some improvements possible for the Mikrotik Wifi. And no, it does not mean the Wave2 yet, nor Wifi6 (AX).

viewtopic.php?f=2&t=145047#p798629

Who is online

Users browsing this forum: No registered users and 18 guests