Re: HD spindown and fs cache writeback
Heather <star@betelgeuse.starshine.org> writes:
> * set noatime on filesystems where you don't care about it anyway
> * make sure you're not syslogging things you don't care about
> * use "noflushd"
> * use "mobile-update"
> * if there's some process named "sync" murder it and see where
> its options are found so you can tweak them
> * Use "hdparm" to advise your disk controller directly that it's
> okay to sleep at more desirable intervals.
I did all this a while back and also shut down every other process I
don't strictly need, but no luck. The drive does spin down
occasionally, but it always spins up again within 30 seconds. I'm
running kernel 2.2.14.
A sample from the syslog:
May 28 23:39:25 jetpack noflushd[297]: Spinning down /dev/hda.
May 28 23:39:45 jetpack noflushd[297]: Spinning up /dev/hda after 0 minutes of spindown.
The fact that it appears is the syslog at all is rather suspicious.
I wouldn't be surprised if that's the problem, that the changed syslog
buffer gets flushed and the disk spins up to accommodate. All my logs
are buffered, but I don't know how to get noflushd to stop syslogging.
That would also explain another curiosity: if something is happening
independently and often enough that the drive never stays spun down
for more than 30 seconds, how is it ever idle long enough to be spun
down in the first place? It seems likely that it's not independent
then, unless something is happening periodically with a period just
longer than the timeout, but that seems rather unlikely.
Elrond <grey.havens@earthling.net> writes:
> for sure WindowMaker tries to access its configuration files every
> three seconds or so.
Wouldn't those files be in the cache then (assuming it's large enough
not to drop them)?
Reply to: