On 2014-09-30 03:20, Sujith Manoharan wrote:

Felix Fietkau wrote:

quoted

This reverts commit 09efc56345be4146ab9fc87a55c837ed5d6ea1ab

I've received reports that this change is decreasing throughput in some
rare conditions on an AR9280 based device

Wasn't the original patch for improving stability ? Won't reverting it
reintroduce the earlier situation ?

The issues that were slightly improved by this patch were fixed in a
different way later. This change has been tested in OpenWrt for 3 weeks now.

If ANI issues/regressions for pre-AR9003 chips are still arising,
I wonder what we have gained by the switch and if the new algorithm
is a good fit for the older chips.

After switching to the new algorithm, a large number of users reported a
very significant increase in stability. Back then, overall stability of
ath9k was a lot worse than it is now, and a much larger number of users
were having trouble with it.

Additionally, the new algorithm is much easier to understand and tune -
the old code (especially the one with both algorithms combined) was an
incomprehensible mess of spaghetti code. I still believe the new
algorithm is better for pre-AR9003 than the old one, just the callback
that applies the settings needs a bit more fine tuning.

The ANI issues that I'm finding these days affect a *much* smaller
number of users, and are less severe than what we've been dealing with
during the time of the switch.

At least for SoC chips, people will report it to OpenWrt. For PC-OEM
cards, users will just say that the card sucks and move on...

For pre-AR9003, most devices used with OpenWrt actually use exactly the
same chips as PC-OEM cards.

- Felix