On Sat, 2006-12-02 at 22:03 -0500, Jon Nettleton wrote: > The kernel has had the scaling governor modules for a while now and I am > not sure why Fedora is still relying on the cpuspeed daemon by default. > If anyone in the know could answer that question I would appreciate it. On my wife's eMachines m6805, running FC6 x86_64, it boots up with ondemand enabled automagically. It seems to be built into the kernel? Its not listed as a module. On my Athlon 64 3000+ desktop, it doesn't work out of the box. Being a desktop, the native PowerNow driver doesn't seem to work but ACPI does. It only has two speeds, 2ghz and 1ghz. I found the cpuspeed demon will flap up and down by default, as its doesn't tune itself to the actual speed difference. I shouldn't have to tune *anything*. I dumped it in favor of loading and enabling ondemand in rc.local, it doesn't seem to have such flapping problems. Another point against userspace is the recent crusade against context switching. A bit ironically, all the userspace polling sends context switching through the roof. In theory, I don't know how its actually implemented, but the kernel governor could run alongside and interact with the kernel scheduler directly, without adding additional context switching and without icky polling.
Attachment:
signature.asc
Description: This is a digitally signed message part
-- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list