I have a routed network that for ordinary web browsing and file transfers worked fine. High speeds (clients are capped to 256k/3Mb u/d) stable downloads and QoS works fine in the main router. All links in 5Ghz.

But I am struggling to get ´real´ traffic flowing without interruptions. I run ROS3.22 or 3.23 (not updated all yet) and the wireless-test package.
On the AP's I started to use the RTS/CTS with Hw.Protection threshold set to 256. That improved VOIP and Skype (L7 in QoS) considerable. (AP's with 20-30 clients). Normal browsing is not noticeable effected.

On my back hauls I has nstreme protocol in use, without framer limit and dynamic size policy.
The raw throughput measured with the bandwidth test tools showed slightly better result then not using nstreme.
CCQ levels showed around 90-95%
Ping time over the link was around 7ms (50b packet size)
Ping time over the link was around 16ms average for a 1500b packet.
But the difference between Hw Frames and Frames reading was huge. About 10 to 20 times higher!

When nstreme was switched off the ping times hardly changed.
The CCQ dropped a bit to around 90%
But the Frames difference was almost back to 1-1!

Since I have been reading somewhere that the frames difference is actually a sign of how many packets are been re-send, and the re-send is a killer for ´real´ time traffic like Voip etc, I think I better drop the nsteme protocol on that link?

Most links are only some kilometres long but on relative low masts (houses) but sufficient Fresnel as far as I can determine (and get..)
To reduce overhead in packages I use short pre-amble.
No web or WPA is used. (mac authentication and the 5Ghz safe guards enough against intruders.)
WMM Suppport is enabled everywhere, but actually not further set (yet) in any router. So basically not in use. Does this WWM support create extra overhead in the packages send over the links?
QoS runs in the main router and some routers through the network. That seems to work fine.

But actually am I looking for some guidance here in how to tweak back hauls to the max to get the lowest latency and lowest package losses so the real time traffic runs fine.
How actually test the links? Is the bandwidth-test a good tool provided you don't run it from the devices making the link?
Is ping a good tool? What size to choose? When should I use ARP ping?
(What size does VOIP packages have? And HTTP? (Skype), HTTPS (Skype) )
If nstreme is used, what is a good framer limit size? Best policy?

I can off course ask clients behind back hauls how their experiences are. But this is the kind of info usually has not a lot of value. The usuall comment you get is "It was good last night" or " It was not as fast as in the beginning." I can't work with that..... To tweak I need fast response/monitoring and results of a setting change. I have several links, some with good signal levels (70dBm or less) some with lesser (around 70-80dBm). Plenty of work!

It would be nice if some of you experts could share some of your extensive experiences so together we could build a sort of guideline and manual that could enrich the wiki.

Rudy

This is very good question. I also want to hear feedback from others.
Please if anyone can support this post.

Thanks.

It is true that the Tx frames vs. Tx Hw Frames is an indicator of retransmits for normal 802.11 operation but there is an additional caveot that makes comparison a little more confusing once the Nstreme protocol is involved.

Beceause the Nstreme protocol can package multiple packets together is one frame (transmission), and/or combine data transmission with the ack frame it is possible that the Tx frames and the Tx Hw Frames could be very different. If you test a link in a non-nstreme state then test in a nstreme case (light load, ie. pings), then test nstreme again (high load) you can get an idea of what is occuring on the actual link and deduce wheither retransmits are occuring or not.

In all my tests with encryption WEP/WPA/WPA2 I have never seen and repeatable loss of bandwidth, so I always encypt my links with the highest and most secure encyption.

I am also interested to see what others have done to as you say 'tweak' the most out of RouterOS.

Cheers

Some things I have found how to optimize for voice.
I run nstreme without any stange configs.
Disable noise immunity. For some reason I found it creates alot of jitter on links. I tested it on a long 43km link, at times the pings shot up to 300ms without any traffic going trough the link.
Lock the speed to the fastest you can get without going below 95% ccq. I start a ping at 30ms timeout though a link and lock it at 6mbps. Then I keep increasing the speed till the ccq drops, and I then set it one setting back. To make sure I then do a speedtest for a while and make sure it still doesn't drop below 95%. Otherwise I set it one more speed setting back.

Something else I have heard from someone working with me. He said that he had a case on a camera link where he got bad video quality, and he saw he had low ccq's on his link. he then disabled WMM and it shot up to over 90%. after that the video feed got alot better.

I usually get a feel for a link's quality by running a both direction TCP speedtest along with a ping. if you do it enough you will get a feeling for good/bad links.

You may also want to not let the radio change the modulation scheme if you are looking for a stable link for voice. i.e. leting only the rate 24 on the AP then the AP will use only 16QAM and probably will hold stable at 100%/100% rx/tx CCQ.

You may also want to not let the radio change the modulation scheme if you are looking for a stable link for voice. i.e. leting only the rate 24 on the AP then the AP will use only 16QAM and probably will hold stable at 100%/100% rx/tx CCQ.

Does this apply to Mikrotik - If yes how?