Hi, On Sat, 9 Jan 2010 15:08:03 +0100, Markus Lindgren wrote: > I did some measurements to check how often my nilfs2 drives are in > standby during one day (24h), during this time only a daily backup > script has been accessing the disks (rsync from some other > directories on /dev/sda1). For comparison I have also included an ext4 > partition on /dev/sdc. Note: there are no other partitions on these > disks so there should basically be no activity. (I am running Ubuntu > 9.10 so an occassional updatedb or similar can be expected, besides my > daily backup script.) > > The nilfs2 settings used during measurement (where I am trying to > force the drives to standby as much as possible, hence, it is not a > typical use of nilfs2): > protection_period = 86400 > nsegments_per_clean = 20 > cleaning_interval = 86400 > (86400 seconds correspond to 24h) > > Results: > nilfs2 /dev/sdb spindown 34 %, count 18, max 2340 sec, spindown > time accumulated 29430 sec > nilfs2 /dev/sdd spindown 47 %, count 26, max 2370 sec, spindown > time accumulated 41220 sec > ext4 /dev/sdc spindown 89 %, count 5, max 53580 sec, spindown time > accumulated 77430 sec > > Same measurement, but this time with no cleanerd running for sdb and > sdd; still the result is in the same range. > nilfs2 /dev/sdb spindown 23 %, count 11, max 2340 sec, spindown time > accumulated 20580 sec > nilfs2 /dev/sdd spindown 47 %, count 23, max 2340 sec, spindown time > accumulated 41190 sec > ext4 /dev/sdc spindown 87 %, count 3, max 55200 sec, spindown time > accumulated 75870 sec > ['count' refers to the number of times the disk has been spun down, > 'max' refers to the maximum time the disk has been in the spun down > state] > > I was quite surprised to see that the nilfs2 disk where quite rarely > in the spun down state, at least when compared to the ext4 disk, > nilfs2 23-47% vs. ext3 ~87%. Also, stopping the cleanerd seamed to > have little effect on disk spinning down. > Could anyone explain what is being done to the nilfs disks during this > time (or provide a reference)? I would have expected the disks to be > spun down much more of the time, for example, on par with the ext4 > disk (~87% spun down). What can I do to reduce disk activity on the > nilfs2 partitions further? For example, should I be using other settings? > > Markus Sorry for the delay. It's certainly curious. I have no clue about what makes the difference. Can you keep track of I/O activity with vmstat command? $ vmstat -p /dev/xxx 1 If reads or writes column regularly increases, something is triggering I/O. Otherwise, something needed by power management may be missing. > ps. I have set the standby timeout on these disk using hdparm; hdparm > -S 30 /dev/xx. hdparm -B did not work for me. > In my measurements I am checking the disks state using 'hdparm -C', > each disk sampled every 30 seconds. The measurements are done using a > bash script. Regards, Ryusuke Konishi -- To unsubscribe from this list: send the line "unsubscribe linux-nilfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
