On Thursday 10 April 2014, Karel Zak wrote: > On Wed, Apr 09, 2014 at 02:39:54PM -0400, Theodore Ts'o wrote: > > On Wed, Apr 09, 2014 at 05:55:57PM +0200, Markus Trippelsdorf wrote: > > > > > The fixed interval is problematic. There are SSD devices out > > > > > there that suffer (their flash cells die out faster) when > > > > > they get trimmed too often. A good rule of thumb is to trim > > > > > once you have written the same amount as you have free space > > > > > on your device. Obviously, that interval varies for every > > > > > user (e.g. it's one week in my case). > > > > > > > > Is "mount -o discard" instead of fstrim interval more or less > > > > bad regarding lifetime? > > > > For those SSD's that have a problem, "mount -o discard" is a > > disaster. Some turn into bricks, others will have a degraded flash > > cells, many will cause extremely degraded performance for other > > processes. > > > > What I usually tell people as far as who ask me for advice is that > > once a week is usually sufficient, especially for most desktop and > > server systems. If you are running an extreme workload which is > > doing a huge number of random writes, then sure, running fstrim > > more frequently, or even using "mount -o discard" might make a lot > > more sense --- especially if you are using PCIe attached flash. > > But in those cases, the system administrator might not want be > > willing to tolerate the random latencies in performance that might > > show up when fstrim is running (for pretty much all SATA and SAS > > attached SSD's out there, they don't yet support queued trim, so > > each trim command requires draining the NCQ queue, which is why > > sending trim commands, whether via "mount -o discard" or via fstrim > > will incur a performance penalty to whatever else might be trying > > to use the disk at the time). > > > > I'll note BTW that even using "fstrim" could potentially brick an > > especially inexpensive/trashy SSD, although the vendor for whose > > drive had been most commonly accused of promulgating those to the > > world is out of business (although there are probably plenty of > > those SSD's still in use in various community distros' audiences.) > > Thanks for the advices. > > I have modified the systemd fstrim.timer (daily -> weekly) and added > some notes to the man page. Thanks, maybe we could add another minor change to "fstrim -a" itself. If possible it would be IMO useful to skip bind mounts to avoid trimming the same filesystem several times in a row like this: $ grep "tmp" /etc/fstab /dev/vg0/tmpdirs /mnt/tmpdirs ext4 acl,user_xattr 1 2 /mnt/tmpdirs/tmp /tmp none bind 0 0 /mnt/tmpdirs/var/tmp /var/tmp none bind 0 0 $ ./fstrim -av /tmp: 392 KiB (401408 bytes) trimmed /var/tmp: 0 B (0 bytes) trimmed /mnt/tmpdirs: 0 B (0 bytes) trimmed Skipping bind mounts should still trim all mounted filesytems. cu, Rudi -- To unsubscribe from this list: send the line "unsubscribe util-linux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
