On Wed, 3 Jan 2001, Michel Dänzer wrote:
> Steven Hanley wrote:
> >
> > On Wed, Jan 03, 2001 at 01:44:35AM +0100, Michel D?nzer wrote:
> > >
> > > Would that be via /proc/sys/vm/bdflush? What do the numbers mean?
> >
> > look in /usr/src/linux/Documentation/sysctl/* notably vm.txt for the bdflush
> > thing.
>
> Thanks!
>
> Does anyone use special bdflush settings for powersaving?
>
This is in my /etc/power/pwrctl-local:
minimum)
$logger -p daemon.info -t pwrctl-local "$1 power $2"
sysctl -w vm.bdflush='60 1000 256 1000 6000 60000 60000 1884 2'
echo 1 >/proc/sys/kernel/powersave-nap
;;
medium)
$logger -p daemon.info -t pwrctl-local "$1 power $2"
echo 0 >/proc/sys/kernel/powersave-nap
sysctl -w vm.bdflush='40 500 128 512 2000 12000 12000 1884 2'
ifconfig eth0 up
;;
maximum)
$logger -p daemon.info -t pwrctl-local "$1 power $2"
$logger -p daemon.info -t pwrctl-local "skipping pwrctl-main"
sysctl -w vm.bdflush='40 500 128 512 1000 6000 6000 1884 2'
;;
(I think the 5th and 6th arguments were the ones that really kept the hd
calm for a while). This is on a MkLinux/Redhat-System, don't know if there is
sysctl on Debian, but it basically just echoes the string to
/proc/sys/vm/bdflush.
I used to have the long update intervals all the time, but they certainly
increase the risk of more serious fs damage (i.e. manual fsck required) in
case of a crash.
Concerning power consumption in MacOS vs. Linux: I surely rather get 4
hours on battery in MacOS, never anything like 5, though (it's a bit
difficult to estimate with that cheapo battery control strip, and I
think I've never been _working_ for 4 hours in MacOS ;).
It seems I now tend to get something below 3 hr in Linux, however,
when I used to get 3.5 when the PB (Lombard) was new. Indeed the typical
current has increased from 1100-1300 mA to 1300-1500.
I could think of 3 explanations:
1. Newer kernels do less aggressive powersaving (sounds irrational,
but maybe the crashes-on-wakeup issues forced Ben et al. to be
more gentle to the pmu).
2. HD is aging (increased friction? HD spin down seems to reduce
consumption only by some 50 mA, though)
3. More memory (192 vs. 64 MB). I hoped this would _reduce_ power
consumption, actually, with more memory for disk cache...
Any ideas?
Cheers,
Derek
