[Discuss] Reducing wear on SSD drives - worth the effort and, if so, how?

[Daniel M Gessel]

TL/DR: Thanks for feedback so far - there's been a wide range of 
thoughts about whether it's worth it; out of curiosity I'm going to 
continue to investigate and experiment.

Boring junk follows:

As to the perspective that persistent storage, generally, is a 
"consumable", I would extend that view to many more components - 
upgrades are the norm (computers become unsupported long before storage 
fails) and clocks may be pushed, some think too far, to the point where 
burnout is likely after the "support lifetime" of many a processor 
(further compounded by the fact that the lifetime management techniques 
may be proprietary). Many users find it worthwhile to overclock, but I 
find it risky. I'm more interested in reducing the rate of consumption. 
Signs above the printers where I used to work reminded us, every time we 
went to pick up a printout, that reducing consumption saves both money 
and reduces waste.

I completely respect the fact that applying the "consumable" label may 
be like an epiphany for some but, while I may be the odd one out, it's a 
very different sort of experience for me.

I do appreciate that a number of folks have mentioned that higher 
quality components have longer lifetimes and I'll keep an eye out when I 
get a replacement.


It doesn't suit everyone's style of communication, but, for better or 
(and often) worse, it's my nature to consider a larger domain when 
pondering a question - I admit I've failed to develop thegood habit of 
being explicit about "adjacent" interests; I'll keep working on it.

Reducing the kind of activity that causes wear on SS storage interests 
me for a number of reasons:

Raspberry Pi users often burnout SD cards and even the cheap ones cost 
almost as much as some of the boards. As these systems may be embedded, 
"zero maintenance" is a windmill that is may be tilted at.Other SBCs may 
have storage on the SOC die or may not be designed to be replaced. 
Phones and tablets don't seem to have easily replaced SS storage.

I also have a system which has a traditional HDD that I run as a server 
but is used intermittently. I've explored wake on LAN/WAN and it doesn't 
seem to support it, so I'm curious about getting the hard drive to stay 
in a power saving state - a wider problem but I think there is some overlap.

Growing the circle a bit more, reducing access to higher levels of the 
memory hierarchy is a general optimization technique: keeping data in 
registers, the on-chip local memory,  minimizing cache misses to avoid 
VRAM accesses, then bus accesses to system RAM, each more expensive in 
terms of time and power consumption.  It's something I don't know much 
about especially when it comes to disks (or CPUs for that matter), so 
I'm curious.


I'm just a tinkererand risk/reward is probably based on a different 
formula than a sysadmin: breaking things, not too badly, is good! It 
creates a learning opportunity. Somebody brought up the mount options 
atime, relatime and noatime. Reading up on this, I prefer noatime and 
I've configured my systems to use it on all storage devices. From what 
I've read, some things may break but probably not HW. As I learn what SW 
depends on it and why, I'll remove it or, because most is open source, 
modify it if I decide I really want to use it.I'm generally interested 
in reducing SW footprint, so finding stuff to get rid of is a bit of a 
background process.

I can refrain from asking tinkerer-y questions if BLU is an sysadmin's 
group or they are otherwise not appropriate for this list.

Thanks,

Dan


On 2022-12-01 16:59, Jerry Feldman wrote:
> I think we need to look at different uses. Certainly, the use of SSDs in
> servers is quite different from consumers. Rich pointed out very
> correctly that they need to be viewed as consumables. I think it is very
> important for home systems to make backups where on a server you might want
> to set up RAID or some auto replication system or file system.
>
> On Thu, Dec 1, 2022 at 4:19 PM Shirley Márquez Dúlcey <mark at buttery.org>
> wrote:
>
>> When SSDs first became available, they were a poor fit for Unix-like file
>> systems unless you made changes because maintaining atime (the time each
>> file was accessed) caused very rapid wear of an SSD. Current distros
>> mitigate that by automatically switching to a modified version of atime for
>> file systems located on an SSD; it only guarantees to show whether the
>> access time is more recent than the most recent change.
>>
>> The very rapid changes to log files can still be an issue in some use
>> cases; again, write-behind caching lowers the impact of that, as the log
>> might be updated multiple times before being written to disk. Systems with
>> extreme workloads might benefit from using a battery-backed RAMdisk for the
>> log files.
>>
>> On Thu, Dec 1, 2022 at 11:16 AM <markw at mohawksoft.com> wrote:
>>
>>> This is a space where "price" or "quality" make a difference.
>>>
>>> A "good" SSD has a lot of extra sectors to map in when it detects a write
>>> error. All done internally to the drive. Better drives do a lot of things
>>> to reduce wear. Some do dedup. Some don't store blocks that are all zero
>>> or blocks that are all ones.
>>>
>>> Its kind of hard to adjust your usage, suffice to say, it is all based on
>>> the amount of change. Individual SSD cells can handle from 3,000 to
>>> 100,000 writes depending on the technology. It is possible to pay twice
>> as
>>> much for a drive that will have 30 times more usable write longevity.
>>>
>>> If your data is largely unchanging, it doesn't matter. If you have a
>>> highly dynamic write environment, go for single level cell NAND flash,
>>> that will last the longest. Find a good enterprise drive that has extra
>>> capacity to remap as cells fail.
>>>
>>>
>>>> Hi all,
>>>>
>>>> The discussion about filesystems got me thinking about whether or not
>>>> it's worth trying to reduce SSD wear on my first system (laptop) to
>> have
>>>> one. It occurred to me that file cloning seems like it could save a few
>>>> writes...
>>>>
>>>> I've heard that some SSDs wear out pretty quickly, but I'm not sure if
>>>> that's real or just rumor and innuendo.
>>>>
>>>> Anyone have thoughts on whether it's worth trying to reduce wear on the
>>>> drive? If so, what kind of changes could I make to my system?
>>>>
>>>> I've installed Ubuntu, which I've been happy with as I'm not much of a
>>>> sysadmin; I know it's resource heavy but I seem to be fine with 16gigs
>>>> of ram.
>>>>
>>>> It's dual boot, but I haven't used windows except when I first got it
>> to
>>>> test; I'll wipe windows if I ever run low on space.
>>>>
>>>>
>>>> Thanks,
>>>>
>>>> Dan
>>>>
>>>> _______________________________________________
>>>> Discuss mailing list
>>>> Discuss at lists.blu.org
>>>> http://lists.blu.org/mailman/listinfo/discuss
>>>>
>>>
>>> _______________________________________________
>>> Discuss mailing list
>>> Discuss at lists.blu.org
>>> http://lists.blu.org/mailman/listinfo/discuss
>>>
>> _______________________________________________
>> Discuss mailing list
>> Discuss at lists.blu.org
>> http://lists.blu.org/mailman/listinfo/discuss
>>
>