I can't say what your issue might have been, but I'd blame a BIOS issue
before the kernel. I saw all kinds of problems when Sandy Bridge first
came out, but they were mostly due to BIOS bugs. Supermicro and Dell
really had to release a lot of BIOS updates. You are right about RHEL
and CentOS having older-than-dirt kernels by default though.
That being said, power management under Linux is not black art. It's
easy to turn off, change, and absolutely does not require a recompile to
alter it. I seriously doubt anyone has ever been harmed by it. My
srcds server hosts all have cpufreq enabled and performance is great.
The general term under Linux is "cpufreq", short for CPU Frequency
scaling. You probably have a module that provides it. By default, if no
CPU scaling module is loaded, the CPU runs at full speed all the time.
The theory is that your CPU is only scaled down to a lower frequency
when it is idle.
To find out if you have cpufreq running, do this:
ls -R /sys/devices/system/cpu/cpufreq
and
lsmod | egrep "cpufreq"
If there is nothing there, then it's not running so stop worrying.
If you do have cpufreq loaded, then there is one thing that I don't like
about the cpufreq defaults. The default ondemand up_threshold is 95%,
which is kinda high.
I have this in my /etc/rc.local to fix it. Adjust as you see fit. 45% is
good for me but 75% might be more reasonable for most people.
echo "Set sane CPUFreq governor threshold."
(sleep 15 && echo -n 45 >
/sys/devices/system/cpu/cpufreq/ondemand/up_threshold) &
Or just disable it by setting your scaling_governor to "performance", or
prevent cpufreq modules from loading.
Suggested reading;
http://www.pantz.org/software/cpufreq/usingcpufreqonlinux.html
http://www.ibm.com/developerworks/linux/library/l-cpufreq-1/index.html
And for Debian and derivatives;
http://technowizah.com/2007/01/debian-how-to-cpu-frequency-management.html
As for Turbo functions, that's all BIOS and static kernel. If you have a
problem there and need to turn it off, the BIOS is the place to go.
I've never seen turbo hurt performance on AMD or Intel, though I've seen
it not working for one reason or another.
1nsane wrote:
A little while back I got a pretty powerful Sandy Bridge server to replace
an older Core 2 Quad and found out it ran SRCDS at worse performance than
the old crappy Core 2 Quad. Apparently the cause was CentOS' power
management not entirely supporting this new CPU and thus bouncing it
between the Turbo Boost Speed and 1.7 GHz.
Fixed the issue by recompiling the kernel with power management options
disabled.
I would recommend checking power saving options. Further thing to look into
is speedstep in the bios.
--
# Jesse Molina
# Mail = je...@opendreams.net
# Page = page-je...@opendreams.net
# Cell = 1.602.323.7608
# Web = http://www.opendreams.net/jesse/
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
https://list.valvesoftware.com/cgi-bin/mailman/listinfo/hlds_linux
