[Cake] Some more cakemq thoughts

[Cake] Some more cakemq thoughts

[Dave Taht]

From: https://docs.google.com/document/d/1tTYBPeaRdCO9AGTGQCpoiuLORQzN_bG3TAkEolJPh28/edit#heading=h.s3q8pyu1s825

...

The kernel has over 2500 places where it can drop packets, many of
which have been instrumented with the  “Drop_reason” facility. No
qdisc has drop_reason encoded into it yet. (Suggested names:
DROP_REASON_{CONGEST, OVERFLOW, FLOOD, SPIKE}).

QOS-MAP mirroring the same syntax as the wifi interface, this
establishes a direct correspondence between cake “tins” and the
settings for the linux wifi qos-map facility.

VLAN-MAP This maps from vlans to tins

NOWASHNQB - cake defaults to nowash allowing the passage of all dscps,
the addition of the NOWASHNQB state means it will wash out everything
except NQB.


-- 
Oct 30: https://netdevconf.info/0x17/news/the-maestro-and-the-music-bof.html
Dave Täht CSO, LibreQos

Re: [Cake] Some more cakemq thoughts

[Sebastian Moeller]

Hi Dave,


> On Aug 31, 2023, at 04:49, Dave Taht via Cake <cake@lists.bufferbloat.net> wrote:
> 
> From: https://docs.google.com/document/d/1tTYBPeaRdCO9AGTGQCpoiuLORQzN_bG3TAkEolJPh28/edit#heading=h.s3q8pyu1s825
> 
> ...
> 
> The kernel has over 2500 places where it can drop packets, many of
> which have been instrumented with the  “Drop_reason” facility. No
> qdisc has drop_reason encoded into it yet. (Suggested names:
> DROP_REASON_{CONGEST, OVERFLOW, FLOOD, SPIKE}).
> 
> QOS-MAP mirroring the same syntax as the wifi interface, this
> establishes a direct correspondence between cake “tins” and the
> settings for the linux wifi qos-map facility.

	[SM] Not a big fan of the qos_map syntax, but I agree that being able to configure the cake DSCP-2-tin mapping from tc seems worthwhile, and not reinventing the wheel with a new way to request that mapping has merits as well (plus one could use the same string for both cake and WiFI), so +1.


> VLAN-MAP This maps from vlans to tins

	[SM] This will IMHO require to switch cake from soft tin-enforcement to strict tin capacity share enforcement (at least under competition) as otherwise the expected priorities are not delivered any more. But I might misunderstand how cake deals with tin's exceeding their share. For normal operation just telling users, do not put greedy traffic into anything > BE seems OK, but for VLANs that will not really fly IMHO.


> NOWASHNQB - cake defaults to nowash allowing the passage of all dscps,
> the addition of the NOWASHNQB state means it will wash out everything
> except NQB.

	[SM] This assumes that NQB will actually work over the existing internet reliably enough to make people switch away from say EF or VA. I think this is wishful thinking (backed in the case of the NQB by underwhelming engineering, and externalization of costs), that said such a feature seems rather benign. Question, many cake keywords come in a negated form e.g. no-ack-filter, so one would expect NOWASHNQB to also come in the form WASHNQB to only wash NQB?

Regards
	Sebastian

> 
> 
> -- 
> Oct 30: https://netdevconf.info/0x17/news/the-maestro-and-the-music-bof.html
> Dave Täht CSO, LibreQos
> _______________________________________________
> Cake mailing list
> Cake@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cake