[Cake] [lede-project/source] Add support for cake qdisc (#72)
dave.taht at gmail.com
Wed Jun 1 11:26:17 EDT 2016
My take on it was doing it in iptables was inefficient, error prone,
and slow. Interfacing to other people's iptables implementations was
and remains especially error prone.
My view has always been that iptables should be used mostly for firewall rules.
I had also had delusions of being able to apply this code and user
facing to other OSes like BSD, which lack iptables, per se'.
I further had delusions of ultimately replacing tc_mirred with
something that could do it as directly part of the ingress qdisc,
where I hoped it would speed up by lot by avoiding a copy.
as for the dscp re-writing issue, by all means, if you want to rewrite
it further than 0, do it with other, custom rules, which eliminates
the api issue jonathon mentioned, after rewriting it to 0, and after
nat is translated. We need better iptables functionality to do dscp
more right in the first place.
Am I the only one whose ISP (comcast) remarks all non-best effort
traffic as background?
On Wed, Jun 1, 2016 at 6:51 AM, Jonathan Morton <chromatix99 at gmail.com> wrote:
>> On 1 Jun, 2016, at 15:25, Benjamin Cronce <bcronce at gmail.com> wrote:
>> 1) Ideally, regardless of platform, should an AQM or scheduler have the responsibility of changing anything other than ECN?
> This was in part my original objection to having the squash/wash feature in Cake.
> The other part is that if we are going to rewrite the rest of the TOS byte (not just the ECN bits), then we should do it properly, which requires a rather substantial extension to the configuration API, even if we only try to cover the most obvious use-cases.
> This would then be a “semi proprietary” DSCP configurator, acting independently of Cake’s core AQM and shaping functions, which would have to be duplicated in other AQMs which had similar aims to Cake. That’s not a good thing, and feeds back into the first point.
> Hence the correct solution is to put DSCP rewriting elsewhere, making it reusable.
> In Linux, doing ingress DSCP rewriting before it hits the ingress qdisc presently requires the rewriter itself to be a qdisc, but this can have Cake as a child qdisc. For the simple “clearing” case iptables can be used instead, as long as Cake is configured to ignore the inbound DSCP using the “besteffort” flag.
> - Jonathan Morton
> Cake mailing list
> Cake at lists.bufferbloat.net
Let's go make home routers and wifi faster! With better software!
More information about the Cake