[Cake] [PATCH net-next] sched: sch_cake: Align QoS treatment to Windows and Zoom
Sebastian Moeller
moeller0 at gmx.de
Fri Jan 10 12:02:18 EST 2025
On 10 January 2025 17:34:45 CET, Dave Taht <dave.taht at gmail.com> wrote:
>actually in re-reading that, I think it leaves nqb in 0 (44, not 45,
>is now VI), which wasn't my intent.
>
>More below...
>
>On Fri, Jan 10, 2025 at 8:24 AM Sebastian Moeller <moeller0 at gmx.de> wrote:
>>
>> Mmmh. off the Linux lists/folks for discussion:
>>
>> can I propose to drop NQB down to BestEffort? The supposed requirements for NQB marking should make these flows thrive in a flow isolating scheduler like cake without any special treatment... moving it to Video or worse Voice really only invites abuse...
>
>I would prefer consistent behavior.
Without proper policing, NQB in Voice, with all/most other traffic in BestEffort a single NQB flow can easily hog up to 50% (I believe this is the cake default) of capacity...
first rule of priority schedulers think twice before exposing a priority scheduler without admission control to the wider internet...
>
>>
>>
>>
>> > On 10. Jan 2025, at 16:55, Toke Høiland-Jørgensen via Cake <cake at lists.bufferbloat.net> wrote:
>> >
>> > From: Dave Taht <dave.taht at gmail.com>
>> >
>> > Cake's diffserv4 mode attempted to follow the IETF webrtc
>> > QoS marking standards, RFC8837.
>> >
>> > It turns out Windows QoS can only use CS0, CS1, CS5, and CS7.
>>
>> Nah, applications apparently can only request a small set of DSCPs, but it is possible to make whole applications use a different DSCP:
>> example for making putty use EF
>>
>> New-NetQosPolicy -Name "putty" -AppPathNameMatchCondition "putty.exe" -PolicyStore ActiveStore -NetworkProfile All -DSCPAction 46
>
>Within an App...
Yepp, that solves zoom's issue, as I am willing to bet default all CS0 zoom does not have fluidity issues due to congesting itself...
>
>>
>>
>> >
>> > Zoom defaults to using CS5 for video and screen sharing traffic.
>>
>> Zoom defaults to using CS0 (my packet captures) unless an adminstrator enables DSCPs, which I assume will only/mostly happen for business accounts.
>
>I have seen it enabled.
Sure, on business accounts that do come with an administrator access to the businesses zoom account... typical home users are CS0 all the way, all the time.
> I have also multiple reports of "stuttering",
>which I am hoping this resolves.
>
>>
>> > Bump CS4, CS5, and NQB to the video tin (2) in diffserv4 mode, for
>> > more bandwidth and lower priority.
>>
>> I do endorse that, but that change will affect all current users of cake as parts of their individual QoS hierarchies.
>
>I think windows users are not as clued.
They are also not that likely to use cake in the first place... so for the convenience on users we do not have we risk annoying experienced users. I am not saying this should stop us but we should consider that, after all this can not easily be overridden via the commandline.
>> >
>> > This also better aligns with how WiFi presently treats CS5 and NQB.
>>
>> Two issues: it is not that default WiFi uses a sane system to begin with, and NQB IMHO should not be elevated in priority unless it is also policed well, something neither WiFi nor cake will do out of the box.
>>
>>
>> >
>> > Signed-off-by: Dave Taht <dave.taht at gmail.com>
>> > Signed-off-by: Toke Høiland-Jørgensen <toke at redhat.com>
>> > ---
>> > net/sched/sch_cake.c | 4 ++--
>> > 1 file changed, 2 insertions(+), 2 deletions(-)
>> >
>> > diff --git a/net/sched/sch_cake.c b/net/sched/sch_cake.c
>> > index 48dd8c88903f..2a9288d4b873 100644
>> > --- a/net/sched/sch_cake.c
>> > +++ b/net/sched/sch_cake.c
>> > @@ -328,8 +328,8 @@ static const u8 diffserv4[] = {
>> > 1, 0, 0, 0, 0, 0, 0, 0,
>> > 2, 0, 2, 0, 2, 0, 2, 0,
>> > 2, 0, 2, 0, 2, 0, 2, 0,
>> > - 3, 0, 2, 0, 2, 0, 2, 0,
>> > - 3, 0, 0, 0, 3, 0, 3, 0,
>> > + 2, 0, 2, 0, 2, 0, 2, 0,
>> > + 2, 0, 0, 0, 2, 0, 3, 0,
>> > 3, 0, 0, 0, 0, 0, 0, 0,
>> > 3, 0, 0, 0, 0, 0, 0, 0,
>> > };
>> > --
>> > 2.47.1
>> >
>> > _______________________________________________
>> > Cake mailing list
>> > Cake at lists.bufferbloat.net
>> > https://lists.bufferbloat.net/listinfo/cake
>>
>
>
--
Sent from my Android device with K-9 Mail. Please excuse my brevity.
More information about the Cake
mailing list