On 12/3/18 3:24 AM, Steve Litt wrote:
> On Sun, 2 Dec 2018 11:10:20 -0800
> Bruce Ferrell <bferrell@???> wrote:
>
>
>> Yeah, this IS one of the issues around flash/SSD storage... They run
>> fast and wear out faster.
> The preceding sentence is true but it's not the whole truth. If one
> uses SSD the way they would spinning rust, that being run it 80% to 90%
> full, with lots of writes, and expects years of service, one will
> likely be disappointed. But there are many situations in which SSD has
> sufficient lifetime.
>
> Let me start with my setup:
>

> =========================================================
> [root@mydesk mnt]# mount | grep sda
> /dev/sda1 on / type ext4 (rw,noatime)
> [root@mydesk mnt]# df -h /
> Filesystem      Size  Used Avail Use% Mounted on
> /dev/sda1       220G   29G  181G  14% /
> [root@mydesk mnt]# mount | grep "/dev/sd"
> /dev/sda1 on / type ext4 (rw,noatime)
> /dev/sdb1 on /boot type ext4 (rw,noatime)
> /dev/sdb7 on /tmp type ext4 (rw,noatime)
> /dev/sdb6 on /var type ext4 (rw,noatime)
> /dev/sdb8 on /run type ext4 (rw,noatime)
> /dev/sdc9 on /home type ext4 (rw,noatime)
> /dev/sdc1 on /s type ext4 (rw,noatime)
> /dev/sdc2 on /d type ext4 (rw,noatime)
> /dev/sdc4 on /classic/a type ext4 (rw,noatime)
> /dev/sdc5 on /classic/b type ext4 (rw,noatime)
> /dev/sdc6 on /classic/c type ext4 (rw,noatime)
> /dev/sdc7 on /home/slitt/mail/Maildir type ext4 (rw,noatime)
> /dev/sdc8 on /scratch type ext4 (rw,noatime)
> /dev/sdc3 on /inst type ext4 (rw,noatime)
> [root@mydesk mnt]#
> =========================================================

>
> * Everything likely to have multitudes of writes under normal operating
> conditions is mounted spinning rust.
>

> * Only 1/5 of the SSD is used, so what few writes there are are
>    distributed across lots of space.

>

> * I delete unneeded stuff and fstrim / every few days, so the SSD
>    doesn't fill up with erased stuff.

>

> * I expect only 4 years life from any drive, spinning rust or SSD. At
>    least half of my disks have blown up within 4 years: That's life. My
>    SSD is currently 4 years old.

>
> The OP's situation differs from mine in one major factor: He has no
> spinning rust to offload writes to. So he'll use all the great
> suggestions in the thread: noatime, put /tmp and logs in RAM
> filesystems, fstrim early and often, and load it exclusively with files
> it's meant to handle (I think the OP wanted an mp3 juke box).
>
> Some people suggested using a USB thumb drive for temp and often
> written files. This is a great idea because you can buy a 64GB thumb
> drive for about $20.00 to $30.00 USD, and just throw it away when it
> breaks. Keep the music on the SSD for speed and reliability, but if the
> music player software happens to write to /tmp, that's on the thumb
> drive that gets replaced every couple years.
>
> An internal 1TB SSD can be had for under $150. External for less than
> $200. If you buy 1TB and be sure to use only 100GB, follow all the tips
> and fstrim every few days, this SSD should last for years. If we assume
> that each song is 5MB, you can hold 20,000 songs in 100GB. If for some
> reason you need to store more than 100GB, well, that's what spinning
> rust is for: Add one.
>
> So it's true, SSDs run fast and wear out faster, but the wear out
> faster part is only if you use them the same way you use spinning rust.
>
>
> SteveT
>
> Steve Litt
> December 2018 featured book: Rapid Learning for the 21st Century
> http://www.troubleshooters.com/rl21
> _______________________________________________
> Dng mailing list
> Dng@???
> https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/dng
>
Everything you say is true Steve... Except for one teeny tiny little thing.

In general people, even engineers, tend to expect storage to behave like storage and not require "special" handling.

At a job I was at a couple of years back, the decision was made to swap SSD for spinning rust in an appliance application that was sold to customers for it's high speed.  The use
case was extreme high speed read/write. It worked REALLY well for that and we got a big boost in perfomance.  After a couple of years in the field, we started to see significantly
disk higher failure rates than we had with spinning rust... And the customers noticed.  Yeah, Oops!

To contrast that, I have 10K rpm spinning rust that has been in use continuously for over a decade, and isn't unusual to see that kind of longevity.

Flash based storage CAN be made to behave, but even then, it will still wear out significantly faster than spinning rust. Generally the failure tends to be catastrophic unless 
utilities to monitor block sparing are used on a regular and on going basis to tell when the unit is approaching that failure point so it can be swapped out before failure.  Those
utilities tend to be specific to the storage, so it's not like "just run smartmon".

Maybe someday there will be more generalized utilities for that type of thing... Not so now.

For now, I know running a Farrari in traffic is a bad idea.  They just don't like that.  So I choose a vehicle best suited to the work load, not the "kewlest" wheels on the road.