[Fork] Asuswrt-Merlin 374.43 LTS releases (Archive)

[john9527]

@fink.nottle A couple of things.....
- on my fork you need to do the 'limits' entry yourself.....that is for the upload/download on the main page take your actual speed and reduce by 10%
- I forget what ASUS has for the default rules for QoS, but seem to remember they were pretty weak. You may need to define your own rule for HTML/port 80 to see any effect on dslreports bufferbloat. I'll take a look and may change the defaults.

It definitely does work when set up correctly. My test goes from a 'C' to an 'A'

 

[cowst]

- re:queuing disciplines...we were able to backport support for codel/fq_codel to the ARM based routers (not possible for the older kernel used on the MIPS routers). It was backported from someone who worked it out for tomato. Not aware anyone has gone to any of the newer disciplines or if that is going to be possible.

So what do we have available on N66U and what can we do about bufferbloat?

 

[john9527]

So what do we have available on N66U and what can we do about bufferbloat?

The new QoS in the beta should work on the N66, you just are limited to using the SFQ queuing discipline. It works, but isn't as good as codel/fq_codel. But it should be a definite improvement over what was there before.

 

[cowst]

The new QoS in the beta should work on the N66, you just are limited to using the SFQ queuing discipline. It works, but isn't as good as codel/fq_codel. But it should be a definite improvement over what was there before.

I'm going to try it.
Is the beta mostly about QoS and if I'm unsatisfied the rest should be mostly like latest stable?

 

[john9527]

I'm going to try it.
Is the beta mostly about QoS and if I'm unsatisfied the rest should be mostly like latest stable?

There's really two main things going on in the beta.....the new QoS (22B4) and DNSCrypt support (added in 22B6). I left them both up in the download area so you can choose which one you would like to try. There's always some misc updates included in both versions, but nothing that should cause any difficulties.

 

[Futuristic]

Whew.....glad to hear it! If it wasn't something simple like that, then it would be ugly to figure out.

Yeah me too! Really appreciate you steering me in the right direction to investigate! Much appreciated.

Also, I have another N66U that runs in AP mode for upstairs and I'm noticing it doesn't resolve the mac address of client very well. Some come through others just show as a mac address. The Main N66U router seems to not suffer from the issue. AP is running 20e9 too and there is no special setup on there either. It does know how many clients are on the network regardless of being an AP, so I was thinking maybe it has its own thing going when in AP mode. I do remember you fixing some issues with the client list in previous versions but maybe only for router mode? I'm not sure how it all works inside, but was just wondering if you had any insight on why this behavior is happening.

Thanks again for all your help! Appreciate it!

 

[cowst]

There's really two main things going on in the beta.....the new QoS (22B4) and DNSCrypt support (added in 22B6). I left them both up in the download area so you can choose which one you would like to try. There's always some misc updates included in both versions, but nothing that should cause any difficulties.

actually taking latest and not using dnscrypt should be probably safe enough to play with qos but have both available, correct?

 

[john9527]

actually taking latest and not using dnscrypt should be probably safe enough to play with qos but have both available, correct?

Yes....If you don't enable DNSCrypt is should stay out of the way

 

[cowst]

Yes....If you don't enable DNSCrypt is should stay out of the way

I am a bit lost after the update: http://www.dslreports.com/speedtest/7494739
I set Traditional QoS to 90% of my nominal speed, and dslreports seems to reach 120mbit like before, but drops at the end and reports a much lower download speed.
Bufferbloat did not move from C, but that might be due to having just a miserable mips N66U

What should I think of these results?

Update: also, I don't see anything in QoS statistics.

Update2: without QoS http://www.dslreports.com/speedtest/7495672

 

[n00b]

Is there anyway to check cpu usage for this? I tried 'top' command via putty but it is showing
CPU: 0.0% usr 0.0% sys 0.0% nic 100% idle 0.0% io 0.0% irq 0.0% sirq

Also, what is the normal temperature for 2.4 GHz - 5 GHz - CPU? I am getting 4xC 5xC 7xC respectively. Is that safe in a long run?
Sorry if this has been asked before, I am new here.

 

[ColinTaylor]

I set Traditional QoS to 90% of my nominal speed, and dslreports seems to reach 120mbit like before, but drops at the end and reports a much lower download speed.
Bufferbloat did not move from C, but that might be due to having just a miserable mips N66U

I found that my N66U didn't have enough CPU power to handle QoS over about 100Mbps. That was the deciding factor in me replacing it with an AC68U.

If you want to test if QoS is working try setting your download speed to 50Mbps and test it again.

 

[sfx2000]

No problem with buffer bloat (not that I had a problem before anyway)

Same here - and this is something completely different...

 

[jrmwvu04]

This is going to be about 800 questions so I apologize in advance... Is there any sort of "QoS rules 101" outline anywhere? I'm toying with it to get it to behave like I want, but I am encountering problems and realizing that it might just be because I'm not setting it up properly.

The upload/download limits are fairly straight forward. I have tinkered with those and mostly understand what I'm doing there.

User defined QoS rules - This is where I have the most questions, particularly surrounding leaving fields empty. If I leave the source IP/MAC empty, I assume it just treats the type of traffic regardless of the source/destination device. Is the reverse true - specifiy the source IP/MAC and leave the port empty, and all traffic to/from a device is treated in a particular way? The "Transferred" fields - I assume it's a min/max, but same questions surrounding blanks. I see 0~512 for the HTTPS for instance. But then file transfer is 512~(blank). So sensibly that means no max, but if you don't want to specifiy the size at all, leave it completely blank? Is that the same as 0~(blank)? And beyond that, how do you come up with sensible values?

User defined priorities - The maximums make sense here, but I'm curious about the minimums. The effects of having the minimums for upload being too high/low aren't clear. Highest Priority packet (ACK/SYN/FIN/RST/ICMP) - I have done some rudimentary research on my own and have some borderline working knowledge of what's going on here, but I haven't decided if there's any reason to deviate from the default values here. But frankly it's a little further down the todo list right at the moment.

Ideas?

 

[john9527]

This is going to be about 800 questions so I apologize in advance... Is there any sort of "QoS rules 101" outline anywhere? I'm toying with it to get it to behave like I want, but I am encountering problems and realizing that it might just be because I'm not setting it up properly.

I always like good questions Makes me think and gives me a chance to double check things. And to give someone a chance to correct me when I'm wrong

User defined QoS rules - This is where I have the most questions, particularly surrounding leaving fields empty.

If I leave the source IP/MAC empty, I assume it just treats the type of traffic regardless of the source/destination device.

True, any traffic of the selected protocol regardless of port.

Is the reverse true - specifiy the source IP/MAC and leave the port empty, and all traffic to/from a device is treated in a particular way?

Right again

The "Transferred" fields - I assume it's a min/max, but same questions surrounding blanks. I see 0~512 for the HTTPS for instance. But then file transfer is 512~(blank). So sensibly that means no max, but if you don't want to specifiy the size at all, leave it completely blank?

You are on a roll

Is that the same as 0~(blank)?

From an end user point of view it's the same, but not the same from the point of view of optimizing iptables/cpu utilization. If you specify limits, it means the QoS iptables chain must be run for each matching packet to check the number of bytes transferred. If you don't specify limits, it means we can set the priority and restore it as the transfer continues without having to run the QoS chain again. My recommendation would be to use transfer limits only where it really makes sense to do so.

And beyond that, how do you come up with sensible values?

Research and tuning to what you want in your environment. For example, I want to throttle back big downloads a bit, but allow full bandwidth to web browsing. A bit of google searching found that the average web site transfer was 1600KB in 2014....so I set my web limit to 0-2048K. (This is what the default rules are trying to do, but their 512K limit seems a bit out of date).

User defined priorities - The maximums make sense here, but I'm curious about the minimums. The effects of having the minimums for upload being too high/low aren't clear.

The min/max values are used in setting up the tc qdisc (Traffic Control Queuing Discipline) classes. Think of the min as the target value you want to achieve, but it's OK to go up to the maximum if you can. It also controls how the available bandwidth is 'shared' between the various tc classes...when one class can give up some bandwidth to help another class meet it's target. So the min/max helps to control which class will be impacted first and by how much as the link gets saturated.

Highest Priority packet (ACK/SYN/FIN/RST/ICMP) - I have done some rudimentary research on my own and have some borderline working knowledge of what's going on here, but I haven't decided if there's any reason to deviate from the default values here. But frankly it's a little further down the todo list right at the moment.

Not much to add here, except when you check the box it's added to the 'Highest' priority tc class. Unchecked, it falls to the default priority setting.

Just one other thing bears mentioning again. The rules are processed in order, first to last. So for example when you are adding both application (specifying only a port) and client (specifying only a MAC), you need to think of which one you want to take precedence.

 

[john9527]

Is that the same as 0~(blank)?

Thinking about this, I just wrote a patch that will convert a rule written this way to a rule without specified limits automatically. Good question

 

[cowst]

I found that my N66U didn't have enough CPU power to handle QoS over about 100Mbps. That was the deciding factor in me replacing it with an AC68U.

If you want to test if QoS is working try setting your download speed to 50Mbps and test it again.

Do you know what happens when you invert upload and download values...?
The above

Otherwise we talk about this:
http://www.dslreports.com/speedtest/7514583

50mbit down doesn't help: http://www.dslreports.com/speedtest/7514632

Actually I got bufferbloat B when accidentally setting down to 5mbit

At these conditions I guess I won't need QoS.

Unless mixing p2p and big downloads affect my browsing, but so far I didn't notice it.

 

[john9527]

Do you know what happens when you invert upload and download values...?
The above

Otherwise we talk about this:
http://www.dslreports.com/speedtest/7514583

50mbit down doesn't help: http://www.dslreports.com/speedtest/7514632

Actually I got bufferbloat B when accidentally setting down to 5mbit

At these conditions I guess I won't need QoS.

Unless mixing p2p and big downloads affect my browsing, but so far I didn't notice it.

Sorry, either you have something set up incorrectly or something is seriously broke in the MIPS netfilter/qdisc. I guarantee the limits aren't reversed.

My results (just taken)

QoS off

QoS Limits at 5 up, 50 down

QoS Limits at 50 up, 5 down

 

[cowst]

Sorry, either you have something set up incorrectly or something is seriously broke in the MIPS netfilter/qdisc. I guarantee the limits aren't reversed.

My results (just taken)

QoS off

QoS Limits at 5 up, 50 down

QoS Limits at 50 up, 5 down

You misunderstood
I inverted them yesterday, now they are correct and the values make sense, although with no other traffic I don't see any benefit with QoS and also the test doesn't show improvements for the C bufferbloat.

 

[john9527]

You misunderstood
I inverted them yesterday, now they are correct and the values make sense, although with no other traffic I don't see any benefit with QoS and also the test doesn't show improvements for the C bufferbloat.

That'll teach me not to respond when I first wake up

Did you look at the detail for the bufferbloat measures as well (click on Download or Upload in the bar chart). Sometime things do improve, but the score stays the same due to a single bad 'spike' measurement.

 

[cowst]

And that'll teach me not to trust my tests just before sleeping