Re: [DNG] noatime by default

[Didier Kryn]

Le 28/08/2017 à 22:10, Alessandro Selli a écrit :
On Mon, 28 Aug 2017 at 12:58:24 +0200
Narcis Garcia <informat...@actiu.net> wrote:
El 28/08/17 a les 11:59, Alessandro Selli ha escrit:
On Sun, 27 Aug 2017 at 17:18:28 -0500
d_pridge <d_pri...@yahoo.com> wrote:
   
Doesn't this affect the expected lifetime for an SSD?
   Little.  AFAIK this used to be a more serious concern on the first
generation of SSDs, because they suffered strongly from write-wear and
because firmware, drivers and filesystems did not support write-levelling.
Today this is much less of a concern.  SSD cells can stand many more write
operations before wearing (not so so called 3D SSD units, however) and
unit's firmware today apply algorithms to write operations that attempt
to spread writes as evenly as possible to cells avoiding impinging too
many times on the same ones.  Which means that, even if you're writing
several times on the same filesystem's blocks (e.g., the FS's log on a
journalled FS), these blocks are mapped to cells spread here and there on
the SSD that are generally different from write operation to another,
transparently to the filesystem's driver and block allocator.  Plus,
SSD-aware filesystems (designed, among other things, to reduce the impact
of write amplification of cells being rewritten) further help prolonging
the unit's life, regardless of how it is used.
   
"SSD-aware filesystems" are flesystems mounted with no atime
   They do more than that:

https://en.wikipedia.org/wiki/Flash_file_system

    JFFS2 and the other filesystems, mentionned in this link are for 
raw flash memory, not for the flash-based disk drives (aka SSD) which 
come with a SATA interface and emulate a rotating disk. You cannot put a 
JFFS2 filesystem on an SSD.

    The last implement wear levelling etc in the firmware, so that they 
can be used with the same filesystems you use on a traditionnal disk. 
But, as was said, it is better if the filesystem is aware of the storage 
being on flash.

    Didier

_______________________________________________
Dng mailing list
Dng@lists.dyne.org
https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/dng