Re: periodic trim for ufs2 ssds

On 12/10/2023 08:49, void wrote:
> On Fri, Dec 08, 2023 at 06:19:51PM -0800, Mark Millard wrote:
>
>> One can sometimes use:
>>
>> # gstat -spod
>>
>> for monitoring alternatives and get an idea if on-the-fly
>> trim is rate limiting activity compared to not having it
>> enabled.
>
> thanks.
> Consumer-grade ssds seem to have sometimes a catastrophic mode of 
> failure where the ssd cannot even be read, by anything.
> I hope that trim might, in the long run, help with delaying that.
> From what I've read so far, trim is considered non-harmful.
>
> On systems with enterprise-grade ssds in a zpool, I have
> (manual, zpool) trim running as a cron job a few minutes before other 
> automatic jobs with heavy I/O. The trim job takes about 3 minutes to run.
> The SSDs are pci not m.2. Hopefully this extends their life to the 
> maximum
> possible extent (together with weekly zpool scrub)
>
Most-common reason for that ("death") in consumer drives is unsolicited 
power failures that result in the internal mapping table getting 
smashed.  There's not much the drive can do about that -- so buy ones 
that don't have that sort of issue (Micron makes some pretty good ones 
at reasonable prices.)

I'm using a decent number of the Micron 5400 series units at present and 
have yet to see any indication of trouble on them, nor any abnormal 
reported wear.  In fact a couple of them in a mirror on a fairly busy 
server where they're serving boot/swap/postfix mail spool, compile and 
similar -- they are still showing, after a full year's use, 100% of wear 
available.

-- 
Karl Denninger
karl@denninger.net
/The Market Ticker/
/[S/MIME encrypted email preferred]/