You can user the laptop-mode-tools for diagnostics, especially lm-profiler for hints who is accessing the disks.
BTW: Linux thecus 2.6.26-1-iop32x #1 Mon Dec 15 19:33:14 UTC 2008 armv5tel GNU/Linux is the kernel I am using, and its sleeping a lot ;-) thecus:/home/tobi# cat /tmp/frostynasstat/hdd_sda_stat Device:/dev/sda State:active LastKnownTemperature:520 EstimatedTemperature:520 TotalSleep:485574 TotalActive:156680 TotalUnknown:0 Spinups:28 DateStamp:"Sun Jan 25 23:09:29 2009" DateLastKnownTemp:"Sun Jan 25 23:09:29 2009" AverageSleep:17341 AverageActive:5595 thecus:/home/tobi# On Sun, 2009-01-25 at 23:10 +0100, Markus Ulbricht wrote: > Hi Matt, > > thanks for the answer. > Since spindown was working properly with debian etch > the noatime mount option was already set. > So I still guess something with hddtemp is now built > into the kernel? > If I send the hdd to spindown with > > /usr/bin/sg_start 0 /dev/sda > > it wakes up after about 4 or 5 minutes and I need > to check out which process might be responsible for that. > > Can give me anyone a hint what I have to check? > > Markus > > Matthew Palmer schrieb: > > On Sat, Jan 24, 2009 at 09:57:29PM +0100, Markus Ulbricht wrote: > > > >> Hi all! > >> > >> Since hdd performance of the Thecus N2100 is not so > >> perfect with debian etch (stable, kernel 2.6.18) I did > >> an upgrade to lenny with kernel 2.6.26 including the > >> DMA patches by Martin Michlmayr. > >> > >> Thanks to Martin the hard drive performance is now > >> much better (173 MB/s cached / 48 MB/s buffered reads). > >> > >> Unfortunately hdd spindown is no longer working. > >> > > > > Oooh, I've been here. My problem turned out to be atime updating; I had a > > process that was reading a file every 30 seconds or so, which then caused > > the atime on that file to be updated and hence the disk was never spinning > > down. Remounting with noatime did the trick. > > > > Assuming that isn't it, it's unlikely that you'll be running enough > > processes to make finding the culprit by killing things one-by-one > > impractical. > > > > > >> Since I am not so familiar with kernel (re)compilation I am not able > >> to create an appropriate kernel needed for tools like atop which is able > >> to report processes responsible for disk i/o. > >> > > > > Rebuilding a kernel isn't hard, but it does take a little while, so I'd go > > with other investigation methods first. > > > > - Matt > > > > > > > > -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
