On 08.05.2010, at 01:59, Björn Smedman <[email protected]>  
wrote:

> On Thu, May 6, 2010 at 6:35 PM, Peter Stuge <[email protected]> wrote:
>> I don't consider ath9k on AR5008 to be at production level for STA
>> with wireless-testing as of a week ago, but AP performance may be
>> different.
>
> I've been running ath9k in a production network (AP mode) for the last
> year or so, but unfortunately have to concur: ath9k is not production
> ready.
>
> The frustrating part is it doesn't really seem to be stabilizing. For
> example the latest "stable" OpenWRT branch (backfire) is currently
> based on compat-wireless from 2010-04-28 and a recent git snapshot of
> hostapd (and I guess there is good reason). That's just not workable
> from a production point of view.
>
> So question is how do we get it to production quality? I think we need
> more focus on stable branches that are maintained and
> stress/interoperability tested, but I'm far from the expert here. What
> is the official plan?
I think what we're doing now is working well enough. The reason why I  
picked the bleeding edge snapshot of compat-wireless for the stable  
branch of OpenWrt is that it fixes a large number of critical bugs.  
However, none of the bugs that I was working on fixing in ath9k were  
regressions.
I think ath9k (and the linux wireless subsystem in general) has had  
very few regressions over time, even in the wireless-testing tree, so  
I see no point in adding more stable branching to the mix.
The compat-wireless snapshot that is in OpenWrt right now has been  
tested on quite a few APs in production use, and while it's not  
perfect, it's in better shape than anything we had before.
Client connections are no longer dropping all the time, the Rx and Tx  
path seems to no longer get stuck, descriptor corruption seems to be  
gone completely - these were all issues that were present in ath9k  
from the beginning, but were fixed recently.
The reason these were not fixed earlier is that other bugs were in the  
way, but that does not imply that the development process is  
ineffective for stabilization.

- Felix 
_______________________________________________
ath9k-devel mailing list
[email protected]
https://lists.ath9k.org/mailman/listinfo/ath9k-devel

Reply via email to