QoS the next generation

[rfdude]

As a method of determining WAN congenstion, you state:
"The controller does this by sending periodic ICMP requests to your WAN gateway and measuring the round trip time. I propose that increased round trip time is a good indication of congestion. The controller sends these requests only when necessary and at a maximum rate of 1 Hz."

How is the WAN gateway defined in Gargoyle for ping purposes?

In my example I'm using a cellular data connection (over HSPA+) with Gargoyle. However, the speed or ping time is quite variable depending on other modems sharing the cell site and which cell site my modem happens to be using. Some sites are still the slower HSPA. Further to the above WAN gateway comment, none is made known to the data modem. I'd probably have to enter something suitable for ping manually?

What do you think? Thanks!

rfdude

[Delphinus]

To compile the images, ensure you have the correct dependencies installed.
On Kubuntu Lucid you'll need:

Code: Select all

build-essential flex libc6-dev zlib1g-dev bison libncurses5-dev

All from apt repository.
and libc6-dev-i386 if you're running a 64bit OS.

[Delphinus]

After compiling, what are the various .img files created? I assume we use gargoyle_1.3.x-x86-generic-combined-jffs2-128k.img as that is what is in the main downloads?

[pbix]

If you are having issues compiling Gargoyle or figuring out what binaries you need to flash your router please start another thread specific to that subject in the forum.

I want to reserve this thread to be about the congestion control and issues you have with it.

[pbix]

rfdude,

I am completely unfamiliar with HSPA+ networks and their nuances. But I cannot think of any reason why the design can not work so I guess you will just have to try it if you want to know.

Some question come to mind.

What are the upload/download speeds you see?

What are the ping times you normally see and how much do they very? If the ping times are varying a lot even when you are not even using the link then it seems the link is always congested and it seems unlikely that QoS can maintain its goals.

What do you want to accomplish with QoS on this link?

The gateway IP address is fixed in /etc/init.d/qos-gargoyle script. You can easily change it there if you think another ping target would be more effective.

[Delphinus]

If you are having issues compiling Gargoyle or figuring out what binaries you need to flash your router please start another thread specific to that subject in the forum.

Are you able to split those posts out into a separate thread?

[Delphinus]

Running this on my alix.

My connection I'm testing this on:
Max downstream (offpeak) 1mbit
Max downstream (onpeak) burst to 1mbit, limited to 500kbit about 30 seconds after a large download is activated.
Max upstream 400-500kbit.

At idle the status is:

Code: Select all

Congestion Control Status
State: WAIT
Link limit: 900 (kbps)
Fair Link limit: 900 (kbps)
Link load: 0 (kbps)
Ping: off
Filtered ping: 13 (ms)
Ping time limit: 54 (ms)

And after I start downloading for example a linux iso from a nearby server, speed hits limit but no change to QOS status - nload is saying:

Code: Select all

Curr: 878.95 kBit/s
Avg: 934.36 kBit/s
Min: 608.13 kBit/s
Max: 1266.13 kBit/s

I notice that from dmesg I'm getting a number of the following errors on router.

Code: Select all

httpd_gargoyle[13347]: segfault at 2524 ip b77c9fd9 sp bfd61e70 error 4 in libcyassl.so.0.0.0[b77bf000+1b000]

[pbix]

Delphinus,
Thanks for posting your results.

Your status screen shows that the controller initialized properly and that you put 1mbs as your link speed. That is correct. Since I cannot see anything wrong let me ask you a few questions.

What QoS classes have you defined for your downlink? Can you send me a screen shot of that?

Was anyone else using your WAN link while you were doing your download?

Please show me a screen shot during your download. When you said it hit the limit what limit did it hit?

During a download as you described if no one else was using the link then the controller should enter the FREE state and the limit should be raised to 1mbps. If this did not happen then something is wrong.

If you can, while the download is occurring please cat /temp/qosmon.status an post the result.

[Delphinus]

Only default QOS classes as downlink, Skype, Voip and BT.
I assume anything not manually specified falls into the default service class "fast"?

Noone else using WAN.

There was no change to QOS status during the download. Hit the ISP limit of just under 1mbit (speed showni nload).

It didn't show as FREE, still only WAIT during max download.

When connection idle:

Code: Select all

root@router:/# cat /tmp/qosmon.status 
State: WAIT
Link limit: 900 (kbps)
Fair Link limit: 900 (kbps)
Link load: 0 (kbps)
Ping: off
Filtered ping: 13 (ms)
Ping time limit: 54 (ms)
ID FFFF, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8004, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8005, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8006, Active  0, Backlog 0, BW bps (filtered): 0.00  

When download running from 1 PC, connection maxed:

Code: Select all

root@router:/# cat /tmp/qosmon.status 
State: WAIT
Link limit: 900 (kbps)
Fair Link limit: 900 (kbps)
Link load: 0 (kbps)
Ping: off
Filtered ping: 13 (ms)
Ping time limit: 54 (ms)
ID FFFF, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8004, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8005, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8006, Active  0, Backlog 0, BW bps (filtered): 0.00   

When download running from 2 different PC's, each getting about 500kBit

Code: Select all

root@router:/# cat /tmp/qosmon.status 
State: WAIT
Link limit: 900 (kbps)
Fair Link limit: 900 (kbps)
Link load: 0 (kbps)
Ping: off
Filtered ping: 13 (ms)
Ping time limit: 54 (ms)
ID FFFF, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8004, Active  0, Backlog 0, BW bps (filtered): 0.00     
ID 8005, Active  0, Backlog 4, BW bps (filtered): 0.00     
ID 8006, Active  0, Backlog 0, BW bps (filtered): 0.00  

[pbix]

Delphinus,

Your report indicates a problem in the controller (qosmon). The controller is not seeing any traffic so it cannot work. Since I do not have this problem on my WRT54G I am going to need your help and patience to figure it out.

Attached find a new (zipped) version of the qosmon program compiled for the ALIX-MADWIFI platform. Please download and unzip the qosmon executable. Disable the controller and replace the /usr/sbin/qosmon with this version. The re-enable the control and retry your test. Please again send the qosmon.status file captured during the test. The new version will put out some additional details I hope will help me.

Thanks..