MikroTik

We have several APs (433AH), each with 30 - 40 Clients (mainly 411s and 133s)
Since the implementation of WMM, VOIP priority over PPPoE has been fantastic.

On the downside, we have found that while a voice call is in progress through the AP, there is a significant reduction in capacity of the data (background) network from any of the other stations. The reduction is many times greater than the actual bandwidth used for the call

Roughly we see upload throughput halve when another station on the AP has 1 call in progress and it gets worse the more calls.

MT - Are there any variables for WMM that might allow us to fine tune the effects of WMM on our APs?

The wmm parameters (AIFSN CWmin CWmax TXOP..) are fixed in Mk.
Have you tried to disable wmm and enable nstreme-polling?

I'm very interested to WMM but i think it's incompatible (http://forum.mikrotik.com/viewtopic.php ... poe#p93730) with my network:

ata(pppoe_client+nat_router+voip)>cpe (bridge_mode) > ap(pppoe-server)

What's your network design.

wm

I just tried WMM on WRT54g as AP

so my experience was.. 2 clients notebooks both WMM activated

when I was alone to AP testing bandwidth I can get TCP 25mbps.. and when others connected its down to half..

Nstreme is not and option for ged as there is no equivalent of WMM's EDCA/HCCA functionality for Nstreme yet.. Hopefully Mikrotik add this and allow tuning of class priorities as it will no doubt be better than standard 802.11



Regards,



Andrew

Gednz, what's your net design?


WMM is applied on transmitted packets (ON BOTH DIRECTIONS):

1)from AP to client AND 2) from client to AP.

In the second situation if your configuration is:
roter-ata(pppoe_client+nat_router+voip) > cpe (bridge_mode) > ap(pppoe-server)
the wmm it will not work because the cpe is just forwarding PPPoE traffic and inspecting encapsulated IP
is not possible, as packets can be encrypted and compressed.

http://wiki.mikrotik.com/wiki/WMM

In this situation is better nstreme because wmm will not work correctly.


The wmm is native for Lan connection and is incompatibility for most Wan connections.


WM

You can mark the VoIP packets and forward the marks, it will change the priority on a packet by packet basis.

OK, You can mark the VoIP packets and forward the marks..... but
In this situation:
router-ata(pppoe_client+nat_router+voip) > cpe (bridge_mode) > ap(pppoe-server),
the wmm it will not work because the traffic through the cpe is encrypted and compressed.
the cpe it will not read the mark and it will not change the priority.

wmillo that is correct, but if you are running the PPPoE client on your Mikrotik CPE you will be able to mark the packets. Come to think of it, if you are running a decent router with built in ATA it should be able to mark the PPPoE packets with DSCP.

We are getting far away from the OP's question, hopefully someone from Mikrotik can chime in and answer the original question.

Our WMM is definitely working in both directions; under load we can disable the priority marking and instantly destroy the VOIP quality in either direction.

More on our setup:
Voice IP form ATA + Data PPPoE from Router <> RB411 Bridge Client <wifi > RB433AH AP Bridge <> PC Based PPPoE-Server and IP router

We keep our VOIP IP separate from our Data PPPoE to make prioritization easier.

To my knowledge, an Nstream AP is not “aware” of its clients wanting to upload priority packets and therefore cannot give them priority over the air; this is what WMM is for.

Maybe if those WMM parameters were adjustable, we could try tuning it up?

By adjusting WMM parameters you could only improve "background" traffic at the expense of higher latency and smaller throughput for your VOIP. The impact of VOIP on overall throughput of AP is caused by the fact, that VOIP uses small packets and 802.11 networks are very inefficent in handling small packets due to per-packet overhead.

Hi ...

that VOIP uses small packets and 802.11 networks are very inefficent in handling small packets due to per-packet overhead.

My dumb questions: those small packets, which size are they?

As far as I remember packet per second are the same so VOIP uses a big packet as any other protocol but without efficiency ... e.g. a single 1500 (or 2456) bytes has only 128 bytes or so of usefull data.

I PCQ limit avialable bw for clients to 512down/128up, short preamble. With such low bw if I start using RTS-CTS with lets say a 512 bytes trigger will I change the PPS X VOIP packets ratio? I mean, without this forced fragmentation I will have a single 128 bytes (guess) each 2457 byte "RF" packet. With the "forced" fragmentation still I have the same 128 bytes but now on a 512 bytes "RF" packet?

I know that forcing small packets reduce the avialable max speed but since each client already have at least 1/10th of a fixed 6Mbps mode this will make too much difference? (I locked APs to 6Mbps G mode to avoid sucessives changes on TX-RX max speeds. CCQ locked at 100 then. No more than 20 clients per AP)

EDITED: Forget abt RTS/CTS on a 802-11g only environement. After some reading here and there, seems that RTS is valid only in mixed B/G environements. Even if I set it at advanced tab, if there is not a 11b client to request to send, the G mode AP will not use it. The 'b' client listen to AP CCK (exists on g mode anyway) but does not listen to hidden b or close/hidden g stations. Close 'g' ones looks like high noise floor to 'b' stations. Than 'b' client RTS if the hidden station mode is enabled on it. AP halts 'g' clients then and CTS to 'b' client. On another language: protect B mode on b/g AP enable/disable.

Sorry agn for the primary questions.

Regards;

Hi Mplsguy

As an experiment: if we disable WMM we can more than double our "background data" throughput (both ways) while maintaining good quality voip streams (same number of voip streams as a control), of course we have to control the amount of background data so as not to disrupt the voip.

I understand about the "lots of small packets" problem, trouble is wmm just reduces the data capacity too much, compared to not using it.