[Bloat] Getting started with sqm-scripts - latency good, bandwidth decimated

moeller0 moeller0 at gmx.de
Wed Jan 20 12:40:09 EST 2016


Hi Brandon,

> On Jan 20, 2016, at 17:10 , Brandon Applegate <brandon at burn.net> wrote:
> 
> I’m getting more confused as I go on :)
> 
> So I’ve rebooted and the tc classes seem to have come back.  Since this is Ubuntu 12.04 - it doesn NOT have systemd.  I mention this because I see there is a systemd config in the sqm-scripts package.
> 
> I have not added any hooks to run 'sqm start’ - neverthless - I have all the rules seemingly there on a fresh boot.  I also have a new ‘interface’ - ifb4eth0.666.  

	This is ours ;) sqm-scripts needs an intermediary functional device or IFB for short to allow attaching a qdisc to the ingress direction; and the IFB4${IFACE} is the convention we adopted in sqm scripts. If you set ingress shaping to 0 this should not appear at all (or only transiently).

> Since I’ve never messed with tc - I have no idea how/where/what is ‘saving’ these rules and making them persistent.  I’m struggling to use google and grep -ir to see where Ubuntu is saving this.

	I believe that has been resolved in later posts already...

> 
> Furthermore - the scripts seem to be working now.  Slightly embarassing - it could be the result of having mucked with the gentoo script and having two many variables flying around at once.

	So sqm-scripts now works in both direction as you expect it? Preserving latency under load performance for a relatively small bandwidth sacrifice?

> 
> I currently lose a small fraction of my bandwidth - but bufferbloat gets an A on the dslreports speedtest.

	Well, you can always try interative to relax the shaper settings again to recover some of the lost bandwidth. But please note that in the ingress direction the shaper works the better the larger the difference between the artificial bottleneck rate (you set for the shaper) and the true bottleneck rate is; I seem to recall that people typically are sufficiently happy with ingress shaping rates of around 85-90% of ingress line-rate.

Best Regards
	Sebastian

> 
> I’m going to concentrate on understanding how / where these rules are getting made ‘persistent’.




More information about the Bloat mailing list