[Wiltshire] Linux 2.6.28, ondemand governor and nice 19 processes

[Robert]

Hello,

I'm hoping someone will be able to help with a problem I've been 
experiencing ever since I upgraded to 2.6.28

I run Boinc (seti at home, etc) as a background process on my quad core 
server at a nice level of 19 and I'm using the ondemand CPU governor.

Before upgrading to 2.6.28 processes running at nice 19 didn't cause the 
CPU frequency to step up, so it ran at it's lowest clock speed, which 
was perfect because it "saved" power and kept temperatures down.

Since upgrading to 2.6.28, however it seems that this is not the case. 
So running Boinc causes all my CPU's to run at full clock speed all of 
the time, which I don't want - not only because it generates more heat 
and uses more power, but it also seems to interfere with processes 
running at higher nice levels, which I never really had a problem with 
before.

I've tried everything I can think of to get back to the previous 
behaviour, but I'm at a bit of a loss what to search for or where to ask 
about this!  Some (useful?) output below.

As a workaround, I've currently limited the CPU time in the Boinc 
manager to only use 50% of the available CPU's which means that 2 cores 
run at 2.4GHz and two run at 1.6GHz, but this doesn't seem to be the 
right solution and still causes the box to run hotter than if both idle 
at 1.6GHz all the time

Any suggestions appreciated!
Robert.


robert at quad ~ $ cat /proc/version
Linux version 2.6.28.4 (root at quad) (gcc version 4.1.2 (Gentoo 4.1.2 
p1.1)) #6 SMP PREEMPT Sun Feb 8 12:42:20 GMT 2009

robert at quad ~ $ cat /sys/devices/system/cpu/cpu?/cpufreq/scaling_governor
ondemand
ondemand
ondemand
ondemand

robert at quad ~ $ cat 
/sys/devices/system/cpu/cpu?/cpufreq/ondemand/ignore_nice_load
1
1
1
1

robert at quad ~ $ cat /proc/cpuinfo |grep MHz
cpu MHz         : 1600.000
cpu MHz         : 2400.000
cpu MHz         : 1600.000
cpu MHz         : 2400.000