Lost speed to ISP when QoS enabled

My ISP guarantees 75Mbps down, and I was actually getting about 95Mpbs down, the actual speed measured by speednet, dslreports, and some others. However, I was starting to get long pauses in Zoom after weeks of using Zoom without this issue.

I decided to try traffic shaping, so I enabled the QoS, using the defaults. I have not modified any of it. Now, the same speed tests are giving me about 45-50 Mbps down. I am not a heavy user of Internet, as I don’t play games, etc., but I am wondering if I need some reconfiguration of QoS.

Hello Harry - I use QoS with all of the defaults and my daughter uses Zoom (and FaceTime) and all seems OK. No long pauses.

I saw your ISP guarantees 75 Mbps. What are your Download & Upload speeds set for the QOS?

I have it set 90000 down, 6000 up. Keep in mind that these are per the actual speeds detected by the speed tests, not the guaranteed speed by the ISP. Thanks for helping.

[edit] The bufferbloat test, when QoS is set in ipfire, gets perfect scores, whereas when it is disabled, it only gets a C+ for bufferbloat.

I see you went from 95 Mbps to 90 Mbps, but you may want to measure the download/upload a few times and go even lower (maybe 3-5% lower).
https://wiki.ipfire.org/configuration/services/qos#calculate-bandwidth-for-qos

 
If the bufferbloat test is better with QoS turned on, it sounds like QoS is working well for that part!

So I dropped it down to 85000 down, and it still gets lousy speeds (about 80 on dslreports, 90 on speedtest.net). When I disable QoS, I get my 90 down again.

Something is going on with QoS that causes my speed to go down severely.

I should have been clearer on my part. With QoS enabled speed tests do slow down. Speed test should be with QoS disabled.

 
What happened with Zoom when QoS is enabled and the Download Speed of QoS was set to 85000 kbps down? Does Zoom still do a long pause?

So you are saying that “limit certain traffic” is the same as loss of overall speed? Sorry, but that isn’t making sense to me. I would think that the faster speed should more than make up for any lost traffic shaping. But that’s an intuitive supposition.

I have not used Zoom in a few days. I have been focused on some other things.

That is they way it works. If a service is getting buried, then QoS will dedicate a slot or lane (or a chunk of bandwidth) just for that service. So hopefully Zoom won’t get buried with QoS turned on.

I use QoS for buffer bloat and for VOIP. It seems to work very well for me!

Glad to hear it is working well for you. Here, not so much.

I would rather enjoy the rated bandwidth (that 95Mbps down, as measured by speedtest and other sites) of my ISP connection than fiddle around with what seems to be a poorly-implemented default QoS implementation. Losing close to 1/2 of the speed of the connection I am paying for hardly seems worth whatever I might be “gaining” from keeping QoS enabled.

If anyone else has a suggestion for improving the QoS configuration, please say so. But until I can get it “right,” I’d prefer to enjoy the full bandwidth of my connection. I will disable QoS for the time being, and revisit it if someone suggests some tweaks.

Thanks.

How are the bandwidth values, if you configure QOS according to your ISP guarantee ( 75MBit/s )?

So I set the Download speed to 70000 (70Mbps), to allow for the 3-5% (though, since I am already well below the rated speed, I probably don’t need that), then re-enabled QoS. Speedtest.net still showing only about 50Mbps down. :roll_eyes:

[edit]: Sorry, I should have mentioned… I notice that after I start QoS, the page seems to be loading for a very long time (about 1-2 mins?). It does eventually quit loading.

Here is the results of the dslreports.com speed test: http://www.dslreports.com/speedtest/64514240. I did great on bufferbloat, but I got an “F” on speed! :frowning:

Your original post stated

and then you thought of adding QoS into the mix to help resolve the problem. Which leaves a few open questions:

  1. You have not said if this actually helped resolve the issue with Zoom.
  2. You have not shown if Zoom was the problem to start off with, or if anything else on the network was to blame for the lag. Maybe even the ISP…

And now the only thing you are concern with is speed…under QoS.

QoS stands for Quality of Service, not speed. This is obtained in a few ways, with an intended outcome of a better overall performance on all services, and keeping some services useable without crippling another, or have one hijacking the available bandwidth. It does not magically increase the existing line speed to perform beyond it’s capabilities, that’s the ISPs department.

“Speed” is not everything. For example: The open hells gate speed of 380kmh for a Bugatti means absolutely nothing if the tyres have zero grip, thus adding to lack of control. Using an old 1970 Alfa GT 1750 at 180kmh, on the other hand will get you to the destination at a “slow” consistent speed. Which experience do you believe is a less stressful, more relaxing, and overall enjoyable? …and perceived as better performing?

Usually QoS means an overall reduction on claimed “speed”, which is a poor indicator on performance, when the connection has high latency, jitter or what not. The reduction should however not be anything more than say 5%.
Again don’t look at the Mbps from speedtest sites (none of them will give you the same numbers), rather do a real world test by downloading a reasonable sized file and timing it.
Preferably while having others users going nuts on the network, have them do VoIP, Skype video, Zoom, what-have-you, Netflix, with and repeat without QoS. Ensure you cleared the cache between the tests, preferably disabled the proxy for both while doing this.
Now compare those numbers, and listen to what your (ab)users have to say about the performance and perceived responsiveness of their respective services during that download period.
If your users feedback is positive under QoS conditions and negative without, you have all you need to know.

In your case set the downlink speed to 72750 (if it’s a 1:1 bidirectional link then add the same on uplink), start with that and tweak it from there. Ensure that your rules 101->220 (or whatever you have last) reflect this as the Maximum Bandwidth. The guaranteed bandwidth will be considerably less, and will be the ones you tweak.

Btw I have found https://speedof.me/ and https://speedsmart.net/ to be a more reliable indicator than speedtest.net (you can also run automated test over a 12hr period https://testmy.net)

If you still get such below par numbers as dslreports showed, and you are sure it’s not your ISP being muppets, then maybe post screenshots of your QoS configuration, lets see what we are working with here.

3 Likes

No, it is not the case that my only concern is speed. My concern, rather, is that I’ve lost close to 50% of it! I don’t see why a feature like QoS should take such a huge bite out of my overall speed.

As to speednet vs the others, I did, in fact, perform multiple tests on various speed testing sites. All of them were reporting (with QoS enabled) somewhere in the neighborhood of 50Mbps or so. Each individual run also produces slightly different results, and if the site is busy, that can also give poor results.

I have not had the opportunity to use zoom recently. I started this thread because I thought QoS might help solve problems with zoom, but this thread is not about zoom at all really. It is really about why I am losing close to 50% of the speed I normally get. Sorry if that confused you.

Believe me, I can live with 50Mbps down. But there is no reason for me to pay for 75Mbps guaranteed (more like 95 down IRL, though) if I can only enjoy half of that most of the time.

[edit] As I mentioned in a previous post, I am using the “preset” configuration and have not modified it further. So there is no reason to post it.

This is normal. I think mine takes about 3-4 minutes for the page to refresh.

You may have graded your speed too high. Is the download speed set to 75?

The load time on the QoS seems to be resource dependent. I have an old quad core Xeon with 16GB RAM, it takes about 5 seconds to reload, needless to say the old Pentium based lightweight I used before took forever to figure out its supposed to do something. Other lightweight UTM (Unifi/MikroTik/etc) devices behave similar. It’s not something uncommon or unheard of, nor unexpected.

Do I understand you correctly, you are paying for a 95Mbps bi-directional connection, of which the ISP will only guarantee 75Mbps?
If yes, then the settings I suggested are correct. Use the speed you know you are able to obtain, and not what the ISP says you might get if no one else is using the pipe… and you are standing right next to the exchange.

The speedtest sites will respond back with a speed that is approximately 2/3 of the speed you have indicated as max. This is because QoS is limiting the amount being pushed. The rest is kept in reserve for VPN/mail/VoIP/what-have-you. It is behaving as intended. Hence I suggested a real life conditions test by downloading a file, with and without QoS, whilst others are using the network as well.

If you want to see 75Mbps in the speedtest sites, you would have to increase your Down/Up numbers to 111550, or there about. However this totally defeats the object of running QoS.
This would be like wanting a fuel economic car, but moan about the lack of torque, rebore the pistons from the 1L engine to a 2L, and then act all surprised that you suddenly own a gas-guzzler that screams for dear life when you floor it. :smile:

I suggest you have another look at the QoS wiki Jon posted higher up, it will help avoid endless headaches :slight_smile:

1 Like

The way that I expect QoS to operate is such that so much bandwidth as is needed at a moment will be used unless something else with a higher priority needs it.

My point is that there is no reason to deliberately waste available bandwidth. As far as your engine/bore/stroke/turbocharging analogy, I’m not sure if that’s appropriate because I really don’t know that much about ICE’s.

Unless OS design has changed in the past 30 years, and I doubt that basic multitasking approaches have in that time, operating systems don’t just throw away CPU processing because we have limited certain processes to some number of cycles; the CPU simply does other tasks while it has those cycles available and surrenders them when other tasks, with higher priority, have the go. At least, that is what I learned by studying the kernels of the AT&T 3B20 and Amdahl UTS Unix operating systems.

In other words, my expectation is that the full amount of bandwidth available – whether that’s the 75Mbps guaranteed by the ISP, or the 95Mbps in real experience – will be utilized by various tasks unless something else pre-empts use of that portion of the remaining bandwidth. So a network speed test (regardless of which one; they all give me about the same results here) uses, say, html and that may be set to, say, 50% of bandwidth by QoS, but that should not PREVENT it from using more if it is available.

Excess bandwidth cannot be stored like excess cash in a bank account. Once the moment goes by, it is gone forever. Meanwhile, the user is frustrated by what becomes an artificial maximum, rather than a guaranteed mininum.

[edit]: to further clarify what I mean

Hello,
qos is not a panacea it can only work with the values you give it and this should never be as high as the maximum of the internet connection see also rely on what the internet connection can do reliably and take a little less:

http://wiki.ipfire.org/en/configuration/services/qos

You are repeating what has already been said. Yes, I am aware that ipfire’s QoS implementation might not be ideal.

Ipfire’s QoS implementation is not what I expected it to be. I am disappointed because I thought Ipfire was intended for hardware embedding. If so, a lot of people might also be upset that they are tossing away a chunk of the bandwidth they are paying for. I thought I had made that clear, but apparently I haven’t, or you have not read the entire thread.

Not used, available. The bandwidth predetermined by your entry will be made available to whichever service needs it, provided the otherside said service is chatting to can handle it. Slowest number wins.

You keep using speed in connection with QoS as though it’s a speed enhancement service.
Performance and speed, whilst not mutually exclusive, do get handled differently when one of them is predefined and no longer a floating variable. It is also of the utmost importance to understand that the quality of a network service is impacted on by a large number of variables.
Bit rate, jitter, transmission delay, packet losses, throughput, delay variation, response time, crosstalk, etc. As you can see speed aka bandwidth plays only one part in a large array of things.

As I indicated in my original response.
Post your configuration screenshot. Lets see what you have, what we are actually working with, and how we can help you tweak it. You declined this offer as you are using preset. Fair enough.
I asked what your lines actually bandwidth is supposed to be, or if the ISP guaranteed speed is it. No answer from you.
I offered you an alternative by increasing your max download and upload to a higher number that would help you get the speed test result you are looking for. ±111550
Yet you declined and here you are still upset about something.

Don’t like my analogies, sure no problem….
Under QoS there is no waste. It is about creating the best connection with the lowest loss potential, stability, and keeping one service/user from hijacking all available bandwidth.

The QoS preset are predetermined best effort configurations defined by the Down/Up speeds you provide. The system cannot smell what bandwidth the line could potentially, under the best conditions, maybe have available for you… Only ofcourse once all planets have aligned and the correct amount of users have been sacrificed to the silicon gods.

Again, YOU have to provide the correct information. Old school computing, garbage in garbage out.
If you wish to obtain better results, however don’t want to follow suggestions made, and don’t accept the help offered, then your only recourse is to get into the subject matter yourself and manually configure all variables to your heart’s content.

1 Like