"Damon L. Chesser" <damon@damtek.com> wrote in message [🔎] 4797448C.4040000@damtek.com">news:[🔎] 4797448C.4040000@damtek.com...
Kalessin wrote:You should stay with the ondemand governor, you will not see any difference, except on yout electricity bill (and on the environnement too). Anyway, just edit /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor and it should work. (as root : echo "performance"/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor)You may need sysfsutils to set it up at boot time : edit /etc/sysfs.conf and write : devices/system/cpu/cpu0/cpufreq/scaling_governor = performance Best Regards.The folks over in #vmserver on freenode say that the frequency scaling messes with the timing of the vms. I can say from my tests so far, this seems to be true. With ondemand my vms seem very sluggish taking up to 30 sec just to run a single ping command on the local net (not timing out, but just to run one ping and get a response) With performance, they are merely slow. So, I think I really need to switch to userspace and manually switch to full throttle when I need to get vm work done.
I can confirm that this is a problem with vm software in general.I'm not sure I've ever seen that ping problem, but I find a real problem with the virtualized kernel's clock skewing if the host's processor frequency is changed. This is because the client kernel never gets properly informed of this change, so it cannot adjust the clock system properly.