Re: Encountered a bug with a dependency of wondershaper, but I'm unsure which dependency, and how to proceed with submitting a bug report
On Tue, Nov 17, 2020 at 09:15:15PM +0100, Graham Bull wrote:
> I've noticed when I set the same rules within wondershaper on Stable and
> Testing, I get different behavior.
> Stable acts as expected, low latency and able to hit the limits set.
> Testing suffers a lot of latency and I'm only able to reach a fraction of the
> limit set.
>
> When I remove the wondershaper rules everything works as expected (I can max my
> internet connection on both computers).
>
> [...]
>
> Any advice of how to collect more info for debugging purposes or how to proceed
> would be very much appreciated!
Short summary: I suggest to dump wondershaper and use CAKE instead.
Please read on for details.
On typical ISP access links, there are devices on both ends of the link
that introduce high network latency under load
(xDSL modem, DSLAM, cable modem, CMTS).
This can be shown well by the DSLreports speed test:
http://www.dslreports.com/speedtest
Often, these devices use large network buffers and too simple buffer
management, which cannot be changed by the user.
Your only chance to reduce the latency is to set up a traffic shaper
on a device under your control to make it a "choke point".
I guess you know this since you are using wondershaper.
A good solution is the combination of a shaper (e.g. HTB) with fq_codel.
wondershaper does not use fq_codel yet.
Personally, I use the CAKE qdisc since it is easier to set up and comes
with even more improvements over HTB + fq_codel:
man tc-cake
https://www.bufferbloat.net/projects/codel/wiki/Cake/
My router runs the OpenWrt distribution, but CAKE should also work on Debian.
Before giving further advice, I would like to know more about
your network:
- Does the Debian box in question manage your Internet access link, or
- is there a separate router, and are there other devices using it?
- What is the type (xDSL, cable, ...) and speed of your Internet link (down/up)?
Regards,
Mirko
Reply to: