QOS configurations and results

I am revisiting QOS and see if it really makes a difference. I have tried this with other routers but they never seem to improve anything. When I first installed IPFire on the previous machine (187) it actually cut the bndwidth almost half and didn’t improve latency. I’m trying again with 192-testing, but I want some assistance finding my settings and @bloater99 offered to help.

Here is the bandwidth test I used to calculate the bandwidth:

Here is my QoS page in IPFire:

Currently when I turn this on, all traffic stops on red

The speeds set at the top are used to preconfigure the Classes. They aren’t actually used to throttle bandwidth. As such, I like to set them quite high, then adjust the bandwidth in each class.

First, Modify the Downlink to 500000 and Uplink to 20000. Then on the bandwidthsettings page, hit Reset. This will set up sane defaults in all the classes.

Now, go to speedtest.net and run it. Your values will be too high, and you will get close to your ISP rated speeds. What to watch for is the ping values next to the speeds. There will be idle ping, download ping, and upload ping. Your download and upload ping will be much higher than your idle ping. As expected with our speeds set too high. This is just a sanity check before we start tweaking.

To start, we aren’t going to touch any settings except Class 103 Maximum Bandwidth and Class 203 Maximum Bandwidth. These classes are for web traffic. HTTP/S, Port 80/443. Start dropping the Maximum value in 103 and 203 by small increments. For example, drop 103 from ~20000 (whatever it is after you click Reset) to 15000. Drop 203 from ~500000 to 450000. Run speedtest again. Keep an eye on the download and upload ping results. When you see them start to drop to close to the idle ping, you know you’re close. Now switch to Waveform Bufferbloat test and fine-tune. Start dropping in smaller increments until the download and upload latency values are close to +0ms. When you’re there, we’re done with running speedtests.

Now, go through the rest of your classes, both uplink and downlink and use the Maximum 103/203 value you just determined as maximal values for the other classes Feel free to adjust downward in classes you know won’t need high values for. For example, my VPN maximum is only 50000 because that’s as fast as the uplink on my remote end can handle. No point in setting the max any higher.

Once you finish all the Maximum Bandwidth on all the classes, now, start adjusting the Guaranteed Bandwidth so that the sum of each class’s Guaranteed Bandwidth is equal to the Maximum Bandwidth you determined. Use your gut, but I typically load up Class 103/203 with the highest guaranteed bandwidth, then proportionally use up the rest. Leave smallish values for ACKs and VOIP should be on the low side. P2P could be dropped way down if you don’t use it. Default I give smallish values.

The main point is that when you sum up all the Guaranteed results in red0 Uplink classes, it should equal the Maximum Bandwidth of Class 103.

Do the same for imq0 Downlink. Make sure the sum of all the 2** Class’s Guaranteed bandwidths is equal to Class 203’s maximum bandwidth.

Now, run some speed tests again and see if it’s any better.

*I referred to Class 103/203 as Webtraffic. I think it’s actually VPN in the default setting. So adjust Class 104/204 if those are your Webtraffic Classes.

Well do you want me to set it higher since they are about the same?

Here is with QoS:

here is without QOS enabled:

Also I find it odd that the webgui stalls when you start QoS until you navigate to a different page in the web GUI

Download ping is 157. Upload ping is 32. Like I said, this is expected. I’m basically starting here as a baseline. Now you should start dropping the Webtraffic class max values, testing after each change. Keep an eye on the down/upload pings. At some point they will start dropping and eventually become close to or equal to the idle ping (which is at 10ms in the screenshot above). I suggest the following values when dropping down:
For download:
460
450
440

For upload:
19
18
17
16

When you hit that point where the down/up pings are equal to the idle ping, then you can start going back up in even smaller increments, to fine-tune.

For example, if you hit 450 and download ping is 10ms, then try 451, 452, 453 etc, until the download ping starts to climb.

Quicker fix is to just go the the top and click on Services->QoS and it’ll reload the page.

1 Like

I just downloaded Kingdom Come Deliverance 2 (about 80gigs). Here is my QoS graph, followed by the gateway ping graph. Notice how the gateway ping barely climbed during the download. This is a benefit of QoS.


That’s nice that QOS works for you. Currently I either need to change that one wire run that became my responsibility (the one from the j box to the modem) or get a different isp which they all are not good here either.
I couldn’t get any pings to go down near the idle ping.

Then I took a nap and this was the results when I woke up:

I think there is still something wrong, but I will try again another time. They changed all the wires but one a few days ago so I’ll change that one and re try this some other time.

But I never had any real luck with QoS working on this cable internet.

As for downloads, I always get really good speeds when the isp isn’t slow.

Also, my graphs don’t work

Hmph. I can’t see how wiring would have anything to do with your results.

What cpu is powering the IPFire box you are setting QoS up on?

1 Like

The computer part is a supermicro X10SLH-LN6TF server with a 3.5 ghz xeon 8gb ram and a TPLink 2.5Gb card for its WAN. There is nothing wrong with that, I have a stack of 20 of the same thing at SteadFast running a lot of other people’s web servers. So this is not the issue and not even pushing this machine like the ones constantly running at the data center with a 100Gb card for uplink.

I’m just going to change out the 40+ year old coax run as it seems it still changes with outside humidity and temperature. Just not as bad as it was.
Plus its the only wire that hasn’t been replaced since the original 1988 installation.
Now you know the irony of my existence: share owner of data centers that has tons of internet, but at the house a tiny cable connection on a very aged CATV system.