Optimized build for the D-Link DIR-860L

Not sure if Bartvz build is similar to the latest trunk snapshot, r4407 with 4.4.71 kernel, but trunk snapshot seems to be running cake with layer_cake fine for me. I messed with settings, adjusted SQM up and down etc, without a router crash or router non-response issues.

http://www.dslreports.com/speedtest/16820620

  • SQM is controlling bufferbloat well for me.

Running a bunch of torrents at the moment, so we'll see.

Interesting. Another user running a newer kernel version with increased stability. I will update my router to 17.01.2 this weekend and test whether SQM is stable.

Could you please be so kind and share a new optimized version, in order to test all of us the same build, based on the same snapshot?
Or should we test the 17.01.2 official release of Jun 11?

Many thanks to keep this very interesting thread active :slight_smile:

My latest test build with the 4.4.71 kernel running cake crashed after running for 2-3 days without problems. I could compile a new build if there are people who want to test LEDE with my modifications. Just let me know :wink:

Yes, please :slight_smile:

Despite these SQM issues, let's not forget and appreciate how stable the Dir-860l is without SQM. Not a single issue since I flashed 17.01.1 one month ago after I was done playing around with custom builds. I'm going to update to 17.01.2 now. Let's hope my experience will be just as pleasant on that build :slight_smile:

http://imgur.com/4ultKWB

OP updated with a new build. This build is untested so flash at your own risk!
Happy testing!

Many thanks Bartvz for the new build!

SQM still crashes, under high load (I have a nice 100mbps down - 100mbps up fiber connection) the router reboots very often (in my case every 30-60 minutes).

Hi all,

Have you already verified if the crashes are strictly linked to SQM or they happen under every high cpu load situation?

I would like to make same tests using the command described in the link below:
https://wiki.openwrt.org/doc/howto/snippets#generate_100_cpu_load

I tried that command, and it would not crash despite a long time of 100% CPU load. So it is not a high load issue.

It also isn't only affecting SQM btw. Limiting your traffic with QOS scripts also results in the same crashes. So it has something to do with shaping network traffic in general I believe

I agree 100%, both qos-scripts and sqm-scripts really are just covenience scripts to set up traffic shaping with the standard linux tools iptables and ip-route2 (tc). The bits they do additionally are only ever executed at set-up time, when you change settings and save&apply, otherwise both are doing nothing.
I do not want to diminish the issue reports, it is clear that there are issues on the DIR-860L (B revision?) that might even be caused by specific kernel modules like cake (but since qos-script is also affected and does not use cake it seems rather more generic). And sqm-scripts might be a perfect method to re-create the problematic situations, but I am afraid that with the current state of affairs there are no actionable items for sqm-scripts to try to help debug this...

@Axl_Mas, @Mushoz & @moeller0: From what I gather the crashes are related to RCU. The weird thing is that without using SQM QoS the DIR860L is rock solid. So, something throws a monkey wrench but what,where and why? Unfortunately, I do not have a clue how to debug the issue. Hopefully some devs will pay attention to our item on the LEDE Flyspray.
For the people who want to read a bit more about RCU and RCU stalls click the words.

@All, OP updated with a new build. This commit is included which should improve WiFi performance! From my testing 2.4 GHz reception and speeds are better.

1 Like

Please add useful information such as this to the bug tracker, so developers reading the issues might have a nice lead where to look. Chances are it's going to be buried in this topic.

1 Like

It looks like some work is being done here on the switch the DIR-860L uses. :slight_smile:

@blogic identified some issues with the CPU core calibration that would explain a whole number of related issues.

He ported some vendor fixes to LEDE and asked for testing help on the mailing list:
http://lists.infradead.org/pipermail/lede-dev/2017-June/008159.html

3 Likes

This sounds very very promising! Will have a look at this and do extensive testing this weekend :slight_smile:

@Bartvz When you have time, could you please be so kind and release a test image with the new patches applied?

We are ready for an extensive testing session! :innocent:

If @Bartvz is busy I can share an image once I compile it this weekend. Stay tuned :slight_smile:

1 Like

@All, OP updated with a new build containing the two patches from this commit by blogic.
The build flashed successfully but I haven't done extensive testing. Enabling SQM QoS with cake in 3, 2, 1...! :smiley:

3 Likes

Many thanks Bartvz for the build!

I have tried it...bad news, after 33 minutes of hard testing with SQM Cake on I have had a reboot, same problem of the previous builds.