WMM vs QoS

WMM vs QoS

[Steven Pease]

Hi,
I've been tracking down some obscure wireless issues and noticed that
there doesn't seem to be a clean mapping between RFC4594 DSCP classes
and WMM queues.

Specifically, it appears that the highest priority defined for VOIP
traffic (EF), and which seems to be commonly used for voice, maps as
AC_VI (next-highest) rather than AC_VO (highest). Whereas I would've
naively expected AF41 to map as AC_VI and EF to map as AC_VO.

I might be misunderstanding something. However, if this is truly the
case, I'm wondering if there would be any adverse effects to modifying
mac80211 to cause EF packets to get translated into the AC_VO queue?
It seems like this way I might be able to have my cake and eat it too
rather than choosing between correct QoS on only one of L2 or L3.

Thanks.

-- 
- Steven

Re: WMM vs QoS

[Steven Pease]

As a test, I attempted to hardcode mac80211 to hardcode a priority of
6 or 7. I tried several places, but the most correct seemed to be
cfg80211_classify8021d. Despite this I noticed no change in behavior.
I also looked for a place in the ath9k driver, but it looked like it
pretty much just handed the data frames off from mac80211 to the
device firmware.

Any pointers on where I should be looking to affect the QoS attached
to 802.11 frames?

Thanks.

- Steven

On Wed, Mar 23, 2016 at 2:57 AM, Steven Pease <spease@suitabletech.com> wrote:
> Hi,
> I've been tracking down some obscure wireless issues and noticed that
> there doesn't seem to be a clean mapping between RFC4594 DSCP classes
> and WMM queues.
>
> Specifically, it appears that the highest priority defined for VOIP
> traffic (EF), and which seems to be commonly used for voice, maps as
> AC_VI (next-highest) rather than AC_VO (highest). Whereas I would've
> naively expected AF41 to map as AC_VI and EF to map as AC_VO.
>
> I might be misunderstanding something. However, if this is truly the
> case, I'm wondering if there would be any adverse effects to modifying
> mac80211 to cause EF packets to get translated into the AC_VO queue?
> It seems like this way I might be able to have my cake and eat it too
> rather than choosing between correct QoS on only one of L2 or L3.
>
> Thanks.
>
> --
> - Steven



-- 
- Steven