How reliable are SSDs in Linux systems?

[paulca]

What happens in case of a warranty claim? Will I get a replacement from a third-party manufacturer, or do they have NOS stock?

Based on "old metal drives". Assuming they are outside the window of "return to retailer".

In the past I have had 2 drives fail inside warranty.  The process isn't made easy for you.  They require you install their diagnostics software, often requiring a USB or CD boot image.  For a Maxtor we had to "Zero the drive" which took 4 hours.  Then we had to run the diagnostics program which took another hour.  Then we had to copy the various fault codes and batch codes etc. etc. into a form on their website.

Finally they accepted the claim.  In the case of a Maxtor they arranged the "Realtime Swap" courier at my cost.
In the other case, can't remember the brand, I had to pay shipping to return it and the replacement arrived via normal post 5 days later.

If you are an enterprise customer or have an enterprise warranty, you might find a van can be at your door that day with a swap out.

EDIT:  I am told that any "Data protection and recovery" warranties are BS.  They will basically just run a bog standard data recovery tool on it and send you what it returns.  They will not be going into any kind of depth you would get from a specialist for several thousand dollars per drive.

[shapirus]

I propose your system has issues.  Something like 150 tabs open in Chrome/Firefox?

No issues. But yeah, I keep a lot of "see it later" tabs open. So partially a usage pattern thing. I'll rather pay a little extra for the memory sticks than worry about cleaning up my tabs that I will need the next day after I closed them. I do use the tab discard feature, though, to save some cpu and memory on inactive tabs.

Other than that, Telegram is using 2.7 GB. Viber is 750 MB. KDE is actually not that heavy: 500+ MB for plasmashell, 300+ MB for kwin, along with Xorg that'll be under 1.5 GB total. That's the memory that is actually used, not the "virtual" allocated value.

Hell with it, I cannot change the way they write software these days, but I still need to use it, so I choose to pay once for the extra memory and forget about it.

BTW, the extra memory will also be utilized for page cache (file cache), which speeds things up and in theory can reduce the amount of physical writes to the disk, so that's another reason to have it.

Speaking of Firefox, I've just tried to kill the one that I use for work to see how big it was, and that alone was using double the size of total used memory in your example, over 6 GB. There's not much in it at all: google workspace chat, some google docs, jira, some github pages etc. Modern web sites built with the amazingly poorly made, bloated and inefficient frontend frameworks, are extremely heavy on resources, both RAM and CPU.

p.s. after restarting that Firefox and restoring the previously opened windows and tabs it uses half the previous amount, about 3 GB. Programming skills my ass...

[paulca]

Telegram, from a quick AI query is known for that apparently.  It caches everything in memory.  Note this is a "cross platform" technique most likely.  When a Linux Native application wants to cache stuff it will usually use a memory mapped temp file.  This allows linux kernel to manage the paging and caching and swap etc.
Browser tabs add up in memory quite a bit.

Memory is funny if you are true honest with yourself.  You could use less of it, but it's available, so you use it.  Then you complain that modern software is happy to let you use it and not optimise itself out of it's real market.  Once you are comfortable using it, you get annoyed when you find yourself on a memory constrained system like an old box repurposed.  "Why's it so bloaty?"

It's not that software kept requiring more and more memory.  It's that hardware folks kept expanding it under us and we decided to use it.

Also, not being memory constrained makes coding easier.  Consider something as low level as fetching a value like "Download % complete" from the download background process.  You could create accessors/IPC to query each element/field so that calls to these are quick, small, fast with little to no allocation required.  OR  You could just write one accessor/IPC that returns the whole contents of the download record with all 25 fields and sublists in one monolithic operation.  You only need one of them.  You might need several dozen integration points for the former, more memory constrainted variant.

Thats one example out of hundreds.  Holding more than one context or task in memory is probably the other big one.  Old applications tended to tear parts down and rebuild new ones when switch contexts, modern applications just make them exist all at once so you can context switch yourself.  Like having many different views open on teh same schematic in the same instance of the editor, versus 10-20 years ago you would need to open who new applications for different views.  (Browsers originally had a single window per instance, then multi-windows, now multi-windows, tabs, groups, workspaces etc.

Validation and constraints.  When you are memory limited, even today, you will set "lower" limitations on the length and size of the contents.  This requires more validation, and more work avoiding over runs.  It also creates difficulties later when those constraints need to change for business requirements and that can be painful with databases etc.  So with more memory and performance available those constraints get set higher.  Values are allowed to be longer.  This enabled the user more and lowers the complexity of the software.

p.s. after restarting that Firefox and restoring the previously opened windows and tabs it uses half the previous amount, about 3 GB. Programming skills my ass...

Do not confuse memory leaks with memory use.  Restarting firefox destroys much of its in-memory runtime state.  There is a fairer argument towards how "sloppy" they are with such in memory caches.

One example, in the case of firefox, a lot of it's profile data is stored in an SQLite database, but that file is memory mapped and "in memory accessed". However, it is, because the linux kernel is, "lazy loaded".  So the pages will not be read into RAM until they are accessed.  After running the browser for a few hours it's back to 2Gb.

[shapirus]

Memory is funny if you are true honest with yourself.  You could use less of it, but it's available, so you use it.  Then you complain that modern software is happy to let you use it and not optimise itself out of it's real market.  Once you are comfortable using it, you get annoyed when you find yourself on a memory constrained system like an old box repurposed.  "Why's it so bloaty?"

It's not that software kept requiring more and more memory.  It's that hardware folks kept expanding it under us and we decided to use it.

Well, for all I know, my computer usage patterns haven't changed much in the last decade or so. I kept tabs open for a long time, used messengers (actually more messengers than I use today) back then, even used VMs more often. And yet somehow 16 GB of memory was more than enough up until about 2017 or 2018 when I doubled the amount because I genuinely needed it for some photo processing software, and then, when I upgraded the entire computer, I decided to go with 64 GB, because it wasn't all that expensive, and I knew I wouldn't regret it, which I didn't.

What I do regret is that I did not upgrade my single 500GB SSD to 2x 2TB in RAID1 configuration when I wanted to do it this summer. In the last two weeks prices have risen by at least 50%, sometimes 100%, which sucks, and the configuration I planned then is no longer so attractive.

[Zenith]

than if you had 16Gb of RAM and swap was hardly touched.

A modern linux-based desktop with a web browser or two, some messengers, a regular (as opposed to something specifically designed as lightweight and therefore featureless) DE like KDE will be constantly swapping with only 16 GB of RAM. 16 GB was more or less fine 10-15 years ago, nowadays I would have at least 32 GB, and no less than 64 GB to not worry about swap at all.

I use xfce and Firefox. I find I can do without the extra features and bloat of KDE or Gnome. I have swappiness set very low. I found that 8GB was too little with Firefox, with many tabs open and left on for 24 hours or more. Firefox seemed to use more RAM the longer it was left on. 16GB was fine. About a year ago Firefox was updated and it was much less of a memory hog, so now there's very rarely significant swapping.

Your mileage may vary.

But, it's miserable and can significantly reduce the life of SSDs to have significant swapping, so it makes sense to have more than adequate RAM.

[paulca]

Yea but the tabs don't contain the same thing they did 10 years ago.  Web2.0 and all.  Half of them are actually natively running and natively allocating Javascript files written by spotty teenagers in bedrooms.

Remember that at least 80% of modern software stack is open source and free.  When you make comments like "good programmers - my arse"...  you are referring to many thousands of people.  Which one screwed up?  Where did they screw up?  Can you fix it?  All options are available to  you.

[paulca]

Y'all seriously sound like you need to visit a "Tab Therapy" counselling season.

I  tend to do it on Fridays in work as a ceremony on logging off.  I close all browser tabs.  On Monday I do NOT "Restrore last session"

It has made my life so much better honestly.  I now reuse tabs and maintain some patterns and sense such that I hardly ever even fill the tab bar on a busy week.

EDIT: In work I have 64Gb on Windows 11.  It's usually sitting around 40Gb used because my memory, saved in my browser tabs, is then spent on running 5-8 copies of InteliJ IDE, each consuming at least 2Gb.

[BitBangBro]

No issues. But yeah, I keep a lot of "see it later" tabs open. So partially a usage pattern thing. I'll rather pay a little extra for the memory sticks than worry about cleaning up my tabs that I will need the next day after I closed them. I do use the tab discard feature, though, to save some cpu and memory on inactive tabs.

...

Speaking of Firefox, I've just tried to kill the one that I use for work to see how big it was, and that alone was using double the size of total used memory in your example, over 6 GB. There's not much in it at all: google workspace chat, some google docs, jira, some github pages etc. Modern web sites built with the amazingly poorly made, bloated and inefficient frontend frameworks, are extremely heavy on resources, both RAM and CPU.

p.s. after restarting that Firefox and restoring the previously opened windows and tabs it uses half the previous amount, about 3 GB. Programming skills my ass...

Leaving browser online for long period of time (days etc) is a BAD idea for many reasons.

One is that there are plenty of exploits that use timing attacks (Spectre/Meltdown etc) that are injected into page's content either at the source or added en route ("man-in-the-middle") that were used by Deep State, but have now trickled down as everything else.

Many of these detect user inactivity to start their job. They are CPU intensive (agressive pattern setting/scanning and timing the response etc etc), some are aggressively allocating/dealoccating memory to trigger/detect  various internal operations etc.

They LOVE inactive long-term browser windows/tabs.

[shapirus]

One is that there are plenty of exploits that use timing attacks (Spectre/Meltdown etc)

Spectre/Meltdown have long been mitigated in the kernel (unless you explicitly turn the mitigation code off).
Are there other similar known vulnerabilities still not fixed?

that are injected into page's content either at the source

NoScript handles these, I don't enable scripts beyond those I need/trust. Yes even those can be malicious, sure, it is not unheard of for a trusted source to get compromised. The auto tab discard feature heps those: inactive tabs are deactivated and are effectively unloaded until I click on them again.

or added en route ("man-in-the-middle") that were used by Deep State

Not possible if HTTPS is used, unless the man in the middle's CA cert is installed in the system's trusted CA store.

But I agree, generally, these are all absolutely valid considerations for a regular user.

[Zenith]

I have to migrate several PCs from Windows 10 to Linux Mint and want to use (small) SSDs as system drives.
The old Windows HDDs should remain in the systems so I can continue to access and edit the old documents.

Anyway, in all this fascinating and informative discussion, I think we've gotten away from the main point. Are you further ahead with this?

It seems to me you have all SATA PCs and you want to use SSDs as boot and system drives, but use the data on the Windows HDDs. Look through the reviews of SATA SSDs and generally on the WWW, and pick something that has a good reputation. Don't skimp on the capacity of the SSDs, because you rarely find you've bought more capacity than you need, and needs usually increase.

4 to 8 GB of RAM may be a problem depending on what you are doing. Put the swap partition on the HDDs. This may require Parted or similar, to resize the partitions on the HDDs and create the swap partition. I believe you should be OK accessing the NTFS (or whatever MS uses now) partitions from Linux, but I'd be inclined to copy the documents to Linux partitions.

Backup the existing systems before you start, because things can go wrong.

[BitBangBro]

Spectre/Meltdown have long been mitigated in the kernel (unless you explicitly turn the mitigation code off).

Not even close.
Everyone and their dog know that this can never be fully fixed.
All they are doing is seeking and plugging individual holes in the Swiss cheese as they find them.
And after the discussion if it makes sense or the perfomrance hit would be uncacceptable.

And that's for kernel, which is "thin" by default - not meant to do much processing on its own.
Userspace is totally different story.

[Greybeard]

Anyway, in all this fascinating and informative discussion, I think we've gotten away from the main point. Are you further ahead with this?
It seems to me you have all SATA PCs and you want to use SSDs as boot and system drives, but use the data on the Windows HDDs.

That's exactly right.

[Zucca]

I would just buy a new Samsung Pro, put whatever filesystem Mint tells you by default and move on (with backups).
There is so much you still have to learn in Linux, do not get sidetracked.

Once you are more confident in Linux and maybe your old laptops are finally dead, then in your new system put ZFS or whatever more advanced.

[abeyer]

I would just buy a new Samsung Pro, put whatever filesystem Mint tells you by default and move on (with backups).
There is so much you still have to learn in Linux, do not get sidetracked.

Once you are more confident in Linux and maybe your old laptops are finally dead, then in your new system put ZFS or whatever more advanced.

1000% this. The flexibility and choice you get with linux can be an asset, and you will find a fanboy somewhere pitching the vast benefits of whatever obscure bit of tech you ask about...  but there are parts of that where you will likely want to make the choice to not make a choice and instead stick to a well trodden path. Alternate filesystems are one of those places I'd never advise someone getting started to mess around with, unless your goal is to play with alternate filesystems. If your goal is to have a working computer, the default filesystem your distro ships will be fine to start with, and you can always explore more options later.

On the SSD situation, I don't even think about it in laptops... even dirt cheap flash storage is more reliable and long lived in a portable device than spinning rust ever was due to the mechanical stresses of being toted around. The only difference that's worth considering is that ssds do seem to be more prone to failing suddenly and catastrophically... you won't necessarily hear clicks or see slow downs or get bad block/io errors first, like you often would with hdds. An SSD may just fall over one day and no longer be readable, period. You should be doing backups anyway, so this is just a good excuse to make sure they are in place and working as intended.

[Simmed]

just in
micron has "disbanded" crucial
no more crucial SSD
but i guess 1 could buy micron SSD

[David Hess]

just in
micron has "disbanded" crucial
no more crucial SSD
but i guess 1 could buy micron SSD

Enterprise drives are all very expensive.

The Crucial drives are the only ones I could find that have power loss protection that works.  I have not been able to confirm any other "consumer" drives which are safe if power is unexpectedly lost.

The reason this is important is that the drive must have some kind of power loss protection if it supports idle time scrubbing.  Otherwise power could be lost in the middle of a scrub operation.

[Greybeard]

The Crucial drives are the only ones I could find that have power loss protection that works.

Regarding power loss:
After my Linux Mint test installation on an external USB hdd, I experienced several situations, after starting some programs, the screen simply froze black.
After a long time of waiting, I could do nothing but shut down the PC.

How dangerous is this situation with SSDs vs HDDs?

[shapirus]

Regarding power loss:
After my Linux Mint test installation on an external USB hdd, I experienced several situations, after starting some programs, the screen simply froze black.
After a long time of waiting, I could do nothing but shut down the PC.

How dangerous is this situation with SSDs vs HDDs?

Speaking of the freeze, it doesn't look like it's related to a specific disk type. Appears to be a more general hardware issue, maybe something with the USB bus or the way it interacts with the external drive's controller, or something else. Even if you physically hot unplug the drive when the system is running, it will not crash that way (I think). I'll expect it to drop into the text mode (or keep the GUI running and allow to switch to console) and dump some kernel or userspace panic log to console and then either freeze or be unable to run anything meaningful.

Either way, what you describe isn't a power loss situation, but more like a system crash. For me it's been a non-issue for many years, regardless of the media type (SSD or HDD). Countless power losses, many cases of having to hit the reset button, and yet no data corruption besides, obviously, the pages that weren't yet flushed to disk which is normally up to a few seconds of last writes. I use ext4 everywhere.

[Greybeard]

Either way, what you describe isn't a power loss situation, but more like a system crash.

It's not caused by a power loss, but a freezing situation that I can only escape by switching off the power.
For the SSD/HDD, the result is the same - a hard power loss.

[shapirus]

Either way, what you describe isn't a power loss situation, but more like a system crash.

It's not caused by a power loss, but a freezing situation that I can only escape by switching off the power.
For the SSD/HDD, the result is the same - a hard power loss.

So my personal experience, FWIW, has been that there's been no unexpected data loss on power cut events regardless of the disk type.

Aside from that, you could try resolving it by using the reset button, if it is present, or, there's a better way, if the kernel is still alive and can accept keyboard input, using the "reisub" magic sysrq combination (use search engines to see what exactly it's about). It requires the respective sysctl setting enabled; many systems, following the trendy path of assuming that their users are retarded and must be constrained in their actions as much as possible, have it disabled by default these days:

Code: [Select]

$ sudo sysctl -a|grep sysrq
kernel.sysrq = 1


[bill_c]

Keep it simple and just use EXT4 for now.
Been using EXT4 since 2010 with HHDs and SSDs (on several systems) and never had a problem, it has survived many unplanned power failures and a few computer resets. I accidentally formatted my main drive once and was able to fix it with gparted, no data loss.

Many other file systems are great, but are designed for other use situations and are overkill for desktop use in most cases.

I tried BTRFS 3 times on different hardware and all 3 times it failed and all data was lost. (its been years ago so maybe it is better now).

[Greybeard]

Keep it simple and just use EXT4 for now.

That's my test HDD at the moment (as suggested by the Mint installation):

[ejeffrey]

For anyone else, F2FS is the king.

Unfortunately, Linux Mint does not seem to support F2FS:
https://forums.linuxmint.com/viewtopic.php?t=442648

You really don't need it.  F2FS or another flash optimized filesystem is really desirable for embedded flash devices like SD cards, eMMC, and that sort of thing. For your standard M2 or SATA SSDs (which have controllers opimized for desktop use) it's basically unnecessary and I wouldn't spend any time worrying about it.

99% of linux desktops and servers out there use SSDs, and they do it with ext4 or xfs or one of the other mainstream filesystems.  They work fine, any anyone telling you you "need" something else is just wrong. 

Buy an SSD from a reputable vendor, and use whatever your chosen distribution picks as default, and you will be good to go.  Don't over think it.

[5U4GB]

(list of things)

All the above plus log2ram to deal with the large amount of Linux software that feels it needs to constantly scribble to log files.  In general, iostat and then process-tracing tools are your friends to identify the worst offenders.

[paulca]

If it helps the OP I have had something like 30 SSDs over the past 10 years and I have had 2 failure and 2 replacements.

The first failure was a Silicon Power drive that went read only on a nano-pc with a very light load after only 3 months.  From a pack of 5x256Gb drives 1 failed.

The second was a "random trash drive" that came in a refurbished Dell corp desktop I bought.  I replaced the disk in that machine, but decided to use it as a boot disc for an experiment server.  It also went read only but in a more spectacular fashion with data loss.

So you should avoid the "non-brand" amazon garbage.

You should also never put important data on one drive, or ideally one place.

I actually still have my original SSD I bought maybe 2015?  A 128Gb SanDisk SATA and it is still in a drive bay powered on and worknig, though very low duty.