Date: Tue, 7 Oct 2014 13:03:58 +0200
From: Stanislaw Gruszka <sgruszka@redhat.com>
To: Johannes Berg <johannes@sipsolutions.net>
Cc: Arend van Spriel <arend@broadcom.com>,
	linux-wireless@vger.kernel.org,
	Zdenek Kabelac <zkabelac@redhat.com>
Subject: Re: [PATCH v2] cfg80211: allow to configure dynamic PS timeout
Message-ID: <20141007110356.GA2902@redhat.com> (sfid-20141007_130642_166386_63337020)
References: <1412083114-8426-1-git-send-email-sgruszka@redhat.com>
 <542AB859.5080008@broadcom.com>
 <542AB893.9090201@broadcom.com>
 <20141001092710.GB2011@redhat.com>
 <1412607640.3098.33.camel@jlt4.sipsolutions.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
In-Reply-To: <1412607640.3098.33.camel@jlt4.sipsolutions.net>
Sender: linux-wireless-owner@vger.kernel.org

On Mon, Oct 06, 2014 at 05:00:40PM +0200, Johannes Berg wrote:
> On Wed, 2014-10-01 at 11:27 +0200, Stanislaw Gruszka wrote:
> > Dynamic power save timeout value is suppose to be configurable via
> > wext, but due to iwconfig bug is not possible to set using that tool.
> 
> That's interesting, what's that bug?

https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=532713

Parsing problem. After reading bug report more detailed I found out it
is fixed on pre beta version v30, but it was not released i.e. on Fedora
we use wireless-tools v29.

> > Allow to configure PS timeout via nl80211 - add NL80211_ATTR_PS_TIMEOUT
> > attribute which become timeout stated in ms.
> 
> Why do you want to be able to set it at all though? I remember having
> this discussion years ago, and we said that it wasn't really useful
> since the user has no idea when and why this should be changed. I'm not
> convinced that changed?
> 
> We had to keep the wext for compatibility - maybe that can now be
> removed if you say it's broken - but I'm not sure I see much value in
> adding it (and you're doing nothing to convince me otherwise, so far)

Zdenek (CCed) reported to me 40% download performance degradation when
PS is used. I think this happen because of delay between packets, but
it is not confirmed yet (I did not provide patches to Zdenek for
testing), hence perhaps problem lies somewere else. I can not reproduce
this issue - I have the same download performance with PS on and off,
I have quite bad performance when set dynamic PS timeout value less
than 20ms, with default 100ms things are fine.

I assume we can provide that setting to user space, if it allow to
fixup performance with PS ?

Zdenek, could you test? You need kernel patch:
http://marc.info/?l=linux-wireless&m=141215578711042&w=2
and two iw patches:
http://marc.info/?l=linux-wireless&m=141208338830615&w=2
http://marc.info/?l=linux-wireless&m=141208338330613&w=2

Thanks
Stanislaw