Re: Regression in device ath between BETA5 and RC1?

On Thu, Oct 20, 2005 at 02:32:05PM -0600, M. Warner Losh wrote:
> In message: <4356E65B.6060009_at_polands.org>
>             Doug Poland <doug_at_polands.org> writes:
> : I'm running 6.0-RC1 on a laptop with a Netgear WG511T PCMCIA ethernet 
> : adapter.  Today I cvsup'd from BETA5 to RC1 and have noticed a serious 
> : regression in performance of this card/driver.  Has anyone else noticed 
> : this?  Should I submit a PR?
> 
> Define serious regression in performance.  I'm unaware of any problems
> in this area.  Is there some way you can quantify which change between
> beta5 and rc-1 appears to have caused this?
> 
Sorry in advance for my lack of technical expertise in this area, I may
be using incorrect or misleading terminology...

I'm running multiple SSH sessions from the laptop to other FreeBSD
boxes.  I'll get periods of time (2-12ish) seconds where the SSH
connection is unresponsive.  This can happen quite frequently, as in
every minute or two.  

In an attempt to better quantify what I'm seeing, I've recorded the
results from an 

  # ifconfig ath0 | grep media: | awk '{print $7}' 

command.  This is what I'm seeing...

This sample was over a period of about 1/2 hour.  I wasn't capturing
time stamps at this point...

OFDM/6Mbps
OFDM/48Mbps
OFDM/9Mbps
DS/5.5Mbps
OFDM/6Mbps
OFDM/9Mbps
DS/11Mbps
OFDM/6Mbps
DS/5.5Mbps
OFDM/6Mbps
OFDM/9Mbps
DS/11Mbps
OFDM/12Mbps
OFDM/18Mbps
OFDM/24Mbps
OFDM/36Mbps
OFDM/48Mbps
OFDM/36Mbps
OFDM/48Mbps
OFDM/36Mbps

Then I got smarter and capture date/time...

Thu Oct 20 15:37:10 CDT 2005 (OFDM/48Mbps)
Thu Oct 20 15:37:32 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 15:39:39 CDT 2005 (OFDM/48Mbps)
Thu Oct 20 15:49:15 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 15:57:10 CDT 2005 (OFDM/24Mbps)
Thu Oct 20 16:02:05 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 16:03:01 CDT 2005 (OFDM/48Mbps)
Thu Oct 20 16:07:08 CDT 2005 (OFDM/54Mbps)
Thu Oct 20 16:14:54 CDT 2005 (OFDM/48Mbps)
Thu Oct 20 16:17:02 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 16:18:10 CDT 2005 (OFDM/48Mbps)
Thu Oct 20 16:21:57 CDT 2005 (OFDM/54Mbps)
Thu Oct 20 16:22:28 CDT 2005 (OFDM/48Mbps)
Thu Oct 20 16:42:57 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 16:59:58 CDT 2005 (OFDM/24Mbps)
Thu Oct 20 17:01:10 CDT 2005 (OFDM/18Mbps)
Thu Oct 20 17:02:57 CDT 2005 (DS/11Mbps)
Thu Oct 20 17:03:54 CDT 2005 (OFDM/12Mbps)
Thu Oct 20 17:05:02 CDT 2005 (DS/11Mbps)
Thu Oct 20 17:06:34 CDT 2005 (OFDM/12Mbps)
Thu Oct 20 17:06:44 CDT 2005 (OFDM/18Mbps)
Thu Oct 20 17:06:54 CDT 2005 (OFDM/24Mbps)
Thu Oct 20 17:07:03 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 17:07:53 CDT 2005 (OFDM/18Mbps)
Thu Oct 20 17:07:55 CDT 2005 (DS/11Mbps)
Thu Oct 20 17:08:03 CDT 2005 (OFDM/9Mbps)
Thu Oct 20 17:08:05 CDT 2005 (DS/5.5Mbps)
Thu Oct 20 17:08:32 CDT 2005 (DS/2Mbps)
Thu Oct 20 17:08:38 CDT 2005 (DS/1Mbps)
Thu Oct 20 17:08:44 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 17:08:50 CDT 2005 (OFDM/24Mbps)
Thu Oct 20 17:08:53 CDT 2005 (OFDM/12Mbps)
Thu Oct 20 17:08:55 CDT 2005 (OFDM/9Mbps)
Thu Oct 20 17:08:57 CDT 2005 (DS/5.5Mbps)
Thu Oct 20 17:09:01 CDT 2005 (DS/1Mbps)
Thu Oct 20 17:09:10 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 17:09:20 CDT 2005 (OFDM/24Mbps)
Thu Oct 20 17:09:22 CDT 2005 (OFDM/12Mbps)
Thu Oct 20 17:09:24 CDT 2005 (DS/11Mbps)
Thu Oct 20 17:09:26 CDT 2005 (OFDM/6Mbps)
Thu Oct 20 17:09:28 CDT 2005 (DS/5.5Mbps)
Thu Oct 20 17:09:30 CDT 2005 (DS/1Mbps)
Thu Oct 20 17:09:38 CDT 2005 (OFDM/36Mbps)
Thu Oct 20 17:09:41 CDT 2005 (OFDM/24Mbps)
Thu Oct 20 17:10:01 CDT 2005 (OFDM/18Mbps)


Every time it "renegotiates" a connect speed, it seems that the
connection is "dead".  Also note, the laptop is stationary during these
measurements.  It also appears that the more I put through the network,
the more these "switches" happen.

I was seeing this behavior a lot on 5.2.1, 5.3, and, IIRC, 5.4.
Responsiveness was so poor as to make remote SSH sessions unusable.
When I installed -CURRENT then RELENG_6, then the NIC started performing
beautifully.  Very stable and consistent.  Now, with yesterday's cvsup'd
to RC1, the NIC/driver is performing poorly again.


-- 
Regards,
Doug