5.4-prerelease - hanging under load

[Nate Lawson]

Christian Brueffer wrote:
> On Sun, Apr 03, 2005 at 12:07:14PM -0400, Pierre-Luc Drouin wrote:
> 
>>Christian Brueffer wrote:
>>
>>
>>>On Sat, Apr 02, 2005 at 09:16:48PM -0500, Pierre-Luc Drouin wrote:
>>>
>>>>Since I upgraded from 5.3-stable to 5.4-prerelease, I've noticed that my 
>>>>computer is hanging badly under load. I've a P4 2.53 GHz without 
>>>>hyperthreading (no SMP). I use the same kernel configurations than 
>>>>before. Now when I compile a port for example, the mouse pointer hangs 
>>>>in Fluxbox and Mozilla takes forever (meaning ~5 sec) to refresh the 
>>>>screen. Even vi hangs. I do not see any warning/error message. Is it a 
>>>>known problem with 5.4?
>>>>
>>>
>>>I have experienced something similar, putting the following into rc.conf
>>>worked for me:
>>>
>>>performance_cpu_freq="HIGH"
>>>
>>>- Christian
>>>
>>>
>>
>>Yes, this seams to fix it. I didn't know that I had a laptop? :)
>>
> 
> 
> Good to hear.
> 
> Nate, this regression was introduced during the cpufreq and friends MFC.
> Is switching the performance_cpu_freq default to HIGH the way to go for
> 5.4-RELEASE?
> 
> - Christian
> 

As you can see from etc/defaults/rc.conf on both -current and RELENG_5, 
we don't currently change the frequency at all:

performance_cpu_freq="NONE"             # Online CPU frequency
economy_cpu_freq="NONE"                 # Offline CPU frequency

However, it sounds like his system's BIOS is booting up with a low 
acpi_throttle setting (probably the lowest one, 12.5% on many systems) 
and so he is seeing very slow performance.  (Only the acpi_throttle 
cpufreq driver has been MFCd for the 5.4 release and the others will 
follow the release.)  Initially, I thought it was safest not to even 
touch the frequency but it looks like it is necessary for some systems 
to always force it high by default.

I'll change the default to HIGH so that we always put systems in the 
fastest performance mode by default.  It will be MFCd quickly as well.

Thanks,
-- 
Nate