Bug 777450

  Basically partition alignment and block size would be an issue of SSD-page (smallest read-writable unit) and SSD-block size (smallest deletable unit) which are claimed to be nor more than 4K and 128*4K=512K by the Linux magazine. However for pre-formatted SSDs I have found file system block sizes of 8K to 32K.

I'm not an expert but from my reading I plugged in these settings:

I set the scheduler for each drive (in my system with mixed drive types) that can be done in /etc/init.d/boot.local:

echo noop >/sys/block/sda/queue/scheduler
echo cfq >/sys/block/sdb/queue/scheduler


I put these commands in the Grub2 bootloader command line as that was suggested if the SSD holds the root file system:

elevator=noop
and
rootflags=data=writeback

The noop scheduler seems to avoid a possible short system hangs that I have seen mentioned for the deadline scheduler in desktop use.


After looking at several articles on partitioning I decided to try the simplest option I found that looked workable, using a start sector of 2048 which is a 1 MB offset that is supposed to work with any combination of SSD page size and filesystem block size.


Adding check boxes for the nodiratime, writeback and discard options to the fstab options screen would be nice too.

Whatever is done will be an improvement over the current situation but it is probably far too much to expect to have everything needed for optimal SSD setup done and tested for 12.2.

Having tried to pass rootflags=data=writeback only rootflags=data got into the grub command line. This may be the first thing to fix.

What about this one for openSUSE12.3? that would really be a great new feature highly demanded for modern mobile computing!

use features.opensuse.org to track features. If there are things to fix, file specific bug reports about specific problems.

btw: a checkbox is useless - the system knows ssds are ssds.

 I am not speaking of a single checkbox, though viewing messages like 'formatted SSD compliant' and 'is SSD with block size 4KB' would value like gold to the user. Summary messages like '/home on SSD' disabling Nepomuk auto-indexing for KDE or updatedb-cron would also be very informative to the user; yes it definitely does matter to the final user whether he is installing on SSD.

* auto-select the right block size (usually 4K), 
* noatime, nodiratime, writeback, discard mount options
* NOOP / CFQ scheduling
* data=writeback root flag
* mount /tmp, /var/tmp to tmpfs
* enable write back caching checkbox for sda, sdb, etc.
* auto-select and view the right partition alignment: 1024, 8, etc.
  (this would also be required for tools like confinedrv)

Look what Stanley Miller has written in comment #2:
  Whatever is done will be an improvement over the current situation but it is
probably far too much to expect to have everything needed for optimal SSD setup
done and tested for 12.2.

  Apart from adding certain mount flags like noatime or writeback the overall task would really need an expert. Nonetheless anything being done will be a definite improvement.

I do personally doubt that this should be resolved as feature to vote on. The distribution provider has to take over responsibility so that the hardware of the user does not incur damage, reduced lifetime or suffer from significant performance penalties.

(In reply to comment #7)
>
> * mount /tmp, /var/tmp to tmpfs

One needs to be careful about moving /tmp without some forethought on what it will impact. An example, k3b uses the /tmp for storage of the working image and fails when them space is exhausted making using RAM a problem unless you have a lot of it when doing a DVD. It would be good to have as an option, just not as the default without a warning note or tweaking the defaults for programs that need huge amounts of temp space.

With 64 GB SSDs selling in the $50 (USD) range more and more folks are going to be adding them.

"comment #5"

I see some stuff on SSDs there, no activity on it for two years and only 9 yes votes.

Yast Partitioning give SSD best practice option #311341: https://features.opensuse.org/311341

I put a link to this bug there.