How reliable are SSDs in Linux systems?

[Greybeard]

...enjoy your new life without Microsoft.

Yesssssss! 

[Greybeard]

Check the parametric search engines.
One on he Geizhals is great, for this any many other purposes.
Here is the list of choices that are M.2 modules, use PCIe4 lanes, are over 1.92TB in size and use various forms of TLC cells (no QLC!). Listed in ascending order of €€/TB:
https://geizhals.eu/?cat=hdssd&xf=7139_%3E%3D1.92TB~7185_3D-NAND%20TLC~7185_3D-NAND~7185_3D-NAND%20TLC%20(eTLC)~7185_2D-NAND%20TLC~7525_M.2%20(PCIe)~7527_PCIe%204.0%20x4&pagesize=100&offset=0&sort=r&promode=true&hloc=at&hloc=de&hloc=eu&hloc=pl&hloc=uk

You can tweak choices to suit your needs, ofcourse.
Even if you don't intend to buy from the list, it helps to see what you are looking for.

Thank you, this is a big help for me.
But I failed to tweak your choices, so I switched to "simple filters".
Which are the "reliable memory cell technology" types (I really don't know), which types should I avoid?

Don't use SSDs, based on elcheapo QLC cells. They are uusually not much cheaper (if at all), but their write speed and rewritability SUCKS.

Is this the only type to avoid?


https://geizhals.eu/?cat=hdssd&xf=7177_2.5%22~7185_3D-NAND%20TLC~7185_3D-NAND~7185_3D-NAND%20TLC%20(eTLC)~7185_2D-NAND%20TLC~7525_SATA~8250_1TB%20(960-1024GB)&pagesize=100&offset=0&fcols=7176&sort=r&hloc=at&hloc=de&hloc=eu&hloc=pl&hloc=uk


https://www.kingston.com/en/blog/pc-performance/difference-between-slc-mlc-tlc-3d-nand

[Greybeard]

Which one(s) would you buy?
https://geizhals.eu/?cat=hdssd&xf=7177_2.5%22~7185_3D-NAND%20TLC~7185_3D-NAND%20TLC%20(eTLC)~7185_2D-NAND%20TLC~7193_ab%20500TB~7525_SATA~8250_1TB%20(960-1024GB)&pagesize=100&offset=0&fcols=7176&sort=p&hloc=at&hloc=de&hloc=eu&hloc=pl&hloc=uk

Which manufacturers would you avoid?

[shapirus]

XFS is traditional "old-timer"/oldie-but-goldie

Is it still prone to zeroing out open files on power loss like it was in the good old days? I remember attempting to use XFS for some database servers about 20 years ago, being lured into it by tales of excellent performance, but all I saw was quite consistent data corruption on power loss and no performance benefits whatsoever compared to ext3. After a couple fun episodes of having to restore entire databases from backups I called it quits, switched to ext3 and later ext4 and never had an issue since.

I'm not sure that anything other than the battle-tested mature ext4 is nowadays worth considering as a daily driver, unless one has a solid precise reason to choose something else. At least for a general purpose system, modern SSD drives (nvme and all that) are so ridiculously fast that the FS performance doesn't matter, and any of them, ext4 included, will be more than fast enough. Therefore, reliability and code maturity considerations should take precedence.

[BitBangBro]

I use F2FS routinely on any SSD inistead of EXT4.

Works great. No issues as far as I can remember.

[Marco]

According to the microbenchmarks from the creator of SSDFS, F2FS isn't much better than ext4 for drive lifetime.

Unfortunately SSDFS itself is not yet ready.

[BitBangBro]

I don't care about microbenchmarks and I din't do it for speed.

I did it because F2FS is far more NAND FLASH friendly and it causes much less wear&tear and it puts much less strain on the controller.

[BitBangBro]

According to the microbenchmarks from the creator of SSDFS, F2FS isn't much better than ext4 for drive lifetime.

Unfortunately SSDFS itself is not yet ready.

Didn'tknow about SSDFS. Interesting.

[RoGeorge]

My SSD is a 512GB from 10+ years ago, and still going fine.

The OS (Ubuntu) and the installs are on SSD, temporary directories in RAM, and other long term stored files and pics are kept on normal HDD spinning disks.

Aside from that, but very important, the file system is ZFS (custom Ubuntu install).  ZFS is SSD friendly by design.  In ZFS, the sectors freed (by deleting files) are not overwritten, they are only marked as free, and kept on the disk until all the other sectors were filled.  Same if you edit a file repeatedly, ZFS doesn't write over and over in the same sector, every time it writes the changes in a new sector, living the old sector intact (but marked as free space).

This also makes possible to make instant disk snapshots, and at no extra disk space cost (a snapshot is just a pointer), and you can roll back and forth between snapshots as much as you like, or mount at the same time different snapshots.  For example you can mount a snapshot from last week and another snapshot from an hour ago, and compare them in your file browser as if it were backups.  Or, you can send to an external disk only the changed sectors (incremental backup).

ZFS has many other crazy features, and contrary to all the online rants about ZFS, it worked for me flawlessly so far, since I don't know how many years (5-10 years).

Usually the install is cumbersome, but doable if you search for 'install <preferred_linux> on ZFS'.

Another thing, ZFS is not recommended for machines with very little RAM, so not good for that old laptop from the attic, with only 2GB of RAM.

[BitBangBro]

To the OP:

IIWY I'd skip all the ZFS crowd by a wide margin.
ZFS is the last thing you need if you just need something that works.

It was made for datacenter and similar use. It offers many features that mere mortals will never need.

And it's CPU and RAM heavy.
Also, it's not that great fit on Linux.
(Which is why BcacheFS is rising fast BTW - it is, amongst other things, made to fit Linux kernel like a glove. BUt it is still in betatesting, albeight mature beta)
It makes a science from a simple file access.

IF you just need a filesystem for your M.2 SSD stick or even HDD, skip ZFS, BTRFS, BCACHEFS and similar.

[Infraviolet]

1. Whether they're reliable in usage is due to whether the manufacturer made an error when designing the SSD and its firmware. SSDs have often needed firmware updates or otherwise some models would operate in modes which did excessive unnecessary writing and limited theie lives. As an SSD has no moving parts the lifetime when in use can be longer than the lifetime of an HDD which is in use but for long-term unpowered storage SSDs are very poor. Don't use them for backup drives, use HDDs for that. Infact I'd sooner trust a backup written with separate a copy on each of two fresh USB flash memory sticks than trust a backup to be on a single SSD.

3.Pick whatever capacity you need for all the documents you regularly work with. I run Mint with just an SSD drive, I don't have HDDS in my system (I use HDDs as external backups but not as something permanently connected). I use 1TB, partitioned as 700Mb of EFI, 250GB of root, 650GB of home and 100GB left over unpartitioned for manual overprovisioning. I try to keep atleast 200GB free in home at all times, and root is naturally pretty empty, 190GB free after over a year's use. I use the default ext4 for the root and home partitions.

4.If you're setting up a system with multiple drives then the SSD is for the operating system, the programs("apps") and some of the most frequently accessed files. HDDs store big unchanging media files and such.

5.Probably not reliably, don't expect any such features to work if you need them. But they do have SMART capabilities like on an HDD and gsmartcontrol on linux can show you these SMART reports.

6.All modern operating systems will be about the same here. I would note that on Mint, I would avoid using swap partitions at all, and set the swappiness very very low indeed, the system will be slower and in some circumstances if your RAM is small there is a risk that something like loads of browser tabs open can freeze your PC, but avoiding swapping prolongs SSD life.

First run
sudo apt install --no-install-recommends smartmontools
and
sudo apt install gsmartcontrol

Do "sudo nano /etc/sysctl.conf" and add the line vm.swappinnes=15 at the very bottom . Then restart and confirm ths has worked by using "cat /proc/sys/vm/swappiness" . Check trim is set to be working: " cat /etc/cron.weekly/fstrim ".

Use "sudo smartd -i 108000" and also edit /etc/default/smartmontools to add the line "smartd_opts="--interval=40800"" at the very bottom. Check these have worked by using "systemctl status smartd.service". The idea behind these commands is not to overstrees the SSD by taking SMART readings too frequently. I thought I rememebred having to edit "/etc/smartd.conf" as well, but I checked  that file on my system and can't see any changes within in which look like I made them, so maybe nothing there is needed.

I also make some edits in Firefox and Chrome to make them use RAM for cache rather than using any SSD space for this, see here: https://easylinuxtipsproject.blogspot.com/p/ssd.html#ID9   https://easylinuxtipsproject.blogspot.com/p/ssd.html#ID10

If you run any virtual machines, take all the same actions inside them.

Never let the system enable hibernation, I think Mint is default against this anyway, Linux hibernation causes all manner of problems on many types of PC hardware well beyond just SSDs, but it is particularly bad for SSDs too.

All my notes in point 6 are for my Mint 21 system, there might be some minor changes when doing it for Mint 22.



As one further note, personally I chose a Western Digital "Red" grade 1TB SATA (my machine hasn't got more modern slots) SSD. It is a bit more professional grade than others and is technically designed to be a cache for a NAS. It does not have the weird "give up fast and get on with other things if there is a read error" mode that most NAS drives have, this drive should keep trying for a long time to re-read if it has any trouble as would any normal desktop drive. It has never had a firmware update made available for it, which suggest there were never any errors made when develeoping its firmware. Firmware updates for any SSD can be troublesome on Linux, so I chose one where I knew I would never need to.

There is one big problem with SSD comparison tables, manufacturers sometimes change the internal design of an SSD without changing the model name/number. The same brand of the same model of SSD might use different flash cells (different grades, different manufacturers...) and a different controller than the same model from months earlier.

[RoGeorge]

Which one(s) would you buy?

Most important is to skip consumer grade, and buy enterprise grade instead.

That will be more expensive, but it will worth any penny in the long term run.  Don't go for the largest consumer disk of the same price.  In a few years any disk will seem small anyway.  Always buy small-enterprise grade storage.  They are many times more reliable, and also much faster under intensive use (faster when it matters, not in benchmarks).

If you buy enterprise disks for home use, you can be sure they'll outlive everything else, and never fail.

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.

However, since I have no experience with SSDs, I have the following questions:

1. How reliable are SSDs in terms of lifespan (limited number of write cycles) and long-term data storage (charge loss in the memory cells)?

2. Which parameters in the SSD datasheets should I pay attention to?

3. What SSD storage capacity should I choose for Linux Mint?

4. Should I also install the programs/apps on the SSD?

5. Do SSDs have automatic data security monitoring that warns of potential data loss in time?

6. Is Linux Mint, as an operating system, generally SSD-friendly (with few write operations), or can it be configured accordingly?

1.  Not a problem when used properly.  For long term, 50+ years or so, I'll go for spinning disks, normal HDD no SSD+HDD hybrid.

2.  Enterprise vs. Consumer grade and the years of guaranteed replacement is the most important, then the size.  Other params may be misleading, marketing luring tricks, depends how it was measured, etc.

3.  Linux distributions are small, any nowadays SSD will be more than enough.  A few GB for the OS alone will be more than enough.  It's the rest of the applications, and maybe the home directory, too, that takes most of the space (in case you want to keep those, too, on the SSD).

4.  Yes, why not.  I do that for speed.  They will launch faster.

5.  Yes.  All disks have SMART these days.  Regarding flash memory wear out, all SSDs have internal wear out balancing algorithms.  Some manufacturers, for example Samsung will advice to let a chunk of SSD free of any partition, so the disk firmware can take advantage of that unpartitioned space, and optimize even better the wear-out ballancing.  This measure is not necessary if you use a file system like ZFS.  ZFS is a CoW file system (Copy on Write), which means it never rewrites the same sector (unless all the free space was exhausted).  It is SSD friendly by design.

6.  No.  No Linux is AFAIK (but I didn't check recently).  The only free OS I've tried with a native CoW file system was FreeBSD (ZFS was the default), but FreeBSD is a BSD flavor OS (like MAC), not a Linux OS.  FreeBSD is also a *NIX OS, but it is much harder to use/maintain as a casual desktop, and with a smaller pool of programs than Debian (which is the nowadays main Linux distro).  Ubuntu is a flavor of Debian, and Mint is a flavor of Ubuntu.

Any Linux can be made SSD friendly if you install on a native CoW file system.  There are a few out there, but when you look into each type, you'll see the only viable option is ZFS.  Mature, rock solid, and can do all the enterprise tricks in case you need that.

For desktop use, ZFS is not memory intensive, and certainly not CPU intensive.  That will happen mostly when you enable automatic sector de-duplication, which is nonsense for casual use.  Usually de-duplication makes a difference only in enterprise application, for example when you have thousands of cloned virtual machines (but each VM slightly different), all stored on the same array of disks.

I use virtual machines on my home desktop (that's how I keep time capsules of software toolchains, for example for older devboards I still have, I keep a VM with all the needed installs, and never let it go on the Internet).  Even so, I still didn't enable deduplication, because the disk space taken by 10-20 VMs is not that big (VMs are kept in HDD space not in SSD).

I don't get why all the fanboysm against ZFS.  Linus didn't want to add ZFS in the kernel for licensing issues, so it never became the default file system, because it is tricky to install Linux on ZFS.  Otherwise, from a technical standpoint, ZFS is hard to beat, and it is not as resource hungry as you may be let to believe.

Of course, to be SSD friendly you can use any other CoW file system.
Except, AFAIK there aren't any as mature and as performant as ZFS. 

[shapirus]

Which is why BcacheFS is rising fast BTW - it is, amongst other things, made to fit Linux kernel like a glove. BUt it is still in betatesting, albeight mature beta

Meanwhile, it was recently removed from the mainline linux kernel and is now available only as a dkms module. Not that it makes it unacceptable by definition, but that's something to consider anyway.

[paulca]

Like a lot of things the kind of SSD and how much you spend combined with your usage details will determine how well the SSD performs to your expectations.

Not all SSDs are alike, not by a long shot.

Again like most things it will come down to "Heavier use?", "Pay more".

Any cheap SSD will work fine as data or system drive.  Even a cheap 256Gb SSD from Amazon would be fine.  It won't be the fastest and if you go for a really cheap one it might only last a few years.

If you stick with "Brands", like Sandisk, Samsung, cKingston, etc. etc. you will pay more but the life span and reliability will increaes, along with a warranty.

This includes running a "swap" partition on it...  again however this depends on your usage.  If you are trying to run a modern desktop in 1Gb or RAM and rely on a 4Gb swap file... it will wear the disk faster than if you had 16Gb of RAM and swap was hardly touched.


However.  If you intend the disk to do any heavy lifting, such as large data transfers into the 10s and 100s of GB....  then you will need to reach deeper into your pocket.  Samsung Evo's are the only "consumer" grade SSDs which don't just lag out after the first Gb transfer and tend to have quiet high "total write wearout limits".  They are about twice the price of others though.  EDIT:  Beware the reddit groups like HomeLab.  They will have you backup machines to SSDs.  This is probably the single most expensive backup you can make.  Consider dumping my night backups to SSD.  Roughly 200Gb written nightly.  With a cheap generic SSD having something like a 40Tb total write limit, that disk will last months.  Even a "pro-sumer" driver with 400Tb write limit will be gone in a year.  They will just say, "Yes, well, us wannabe professionals use "Enterprise grade" SSDs which cost about £200 a terrabyte and have write limits 10 to 100 times higher.

So if you want a basic HD for a bog standard Linux desktop/devmachine/server go with any "known" brand vanilla generic "SATA" SSD.

If you want speed, look for an M.2 NVMe drives and study up on bus speeds, lanes, etc.  Hell add two in striped (my Steam drive) which reads at 10Gb/s.

EDIT:  Note:  SDCard != SSD.  An SDCard has pretty much zero protection for data loss and a near digital "Im fine", "xxxxxx" failure mode.  SSDs are a lot more "self aware" and track data write fails over time and with SMART enabled it will start complaining long before you absolutely must replace it.  Usually the disk retains all data fine but will go into "read only" mode as a last resort before failing.

[Zenith]

Which one(s) would you buy?
https://geizhals.eu/?cat=hdssd&xf=7177_2.5%22~7185_3D-NAND%20TLC~7185_3D-NAND%20TLC%20(eTLC)~7185_2D-NAND%20TLC~7193_ab%20500TB~7525_SATA~8250_1TB%20(960-1024GB)&pagesize=100&offset=0&fcols=7176&sort=p&hloc=at&hloc=de&hloc=eu&hloc=pl&hloc=uk

Which manufacturers would you avoid?

Samsung and the big names are the safest choices, but they too put out some "value engineered" products, because there's always a market for cheap. The warranties are worth noting. I'd avoid anything DRAM-less or with QLC for serious use. It's different for an old laptop you only use for car diagnostics or something like that.

I haven't bought an SSD for a couple of years so I'm not up to date with what's on offer now. There are regular reviews on the WWW of individual SSDS, and roundups which include maybe 10. They tend to concentrate on performance, which may be irrelevant. If you are using SATA then that's a limitation, although you'll find any SSD will boot and load applications noticeably faster than an HDD. If you are using NVMe, data transfers start about four times faster than SATA and are probably by now 10 times faster, depending on what your motherboard is capable of and how much you are prepared to pay for the SSD.

For some users, such as professional video editors, handling maybe 100GB of data at a time, performance is important, and they are worried about things like what size of file can be written without slowing data transfer to a crawl. For high capacity top end drives that may well be over 100GB. They have a need and for them it's worth paying for that. Most of us would notice no difference.

I have two PCs I use a lot. One had a Lexar 256GB SSD and a Toshiba 3TB HDD for about four years. The Lexar was getting full so was swapped for a 500GB Crucial MX500. The other has a Sabrent Rocket 1TB NVMe, which was widely recommended a few years back.

As for cheap SSDs:

For years I ran a Raspberry Pi as an always on torrent box and that used a Patriot Burst 480GB SSD via a USB adapter.

I have a third PC I occasionally use for experimenting with Linux versions, or running Win10 for things it's awkward to run in VirtualBox. The SSDs are very cheap ASUS and Integral 128GB.

I have a couple of thin clients I use with SSDs. One's an old 60GB Samsung SSD removed from its case.

None of these things are critical or involve many writes to the SSD.

[paulca]

I think bang in the middle of the "better" side of the consumer market are these:

https://www.amazon.co.uk/Samsung-NVMe2-0-Internal-Solid-MZ-V9S2T0BW/dp/B0DGHB9V34/ref=sr_1_1?sr=8-1

[Greybeard]

1. Whether they're reliable in usage is due to whether the manufacturer made an error when designing the SSD and its firmware. SSDs have often needed firmware updates or otherwise some models would operate in modes which did excessive unnecessary writing and limited theie lives. As an SSD has no moving parts the lifetime when in use can be longer than the lifetime of an HDD which is in use but for long-term unpowered storage SSDs are very poor. Don't use them for backup drives, use HDDs for that. Infact I'd sooner trust a backup written with separate a copy on each of two fresh USB flash memory sticks than trust a backup to be on a single SSD.

3.Pick whatever capacity you need for all the documents you regularly work with. I run Mint with just an SSD drive, I don't have HDDS in my system (I use HDDs as external backups but not as something permanently connected). I use 1TB, partitioned as 700Mb of EFI, 250GB of root, 650GB of home and 100GB left over unpartitioned for manual overprovisioning. I try to keep atleast 200GB free in home at all times, and root is naturally pretty empty, 190GB free after over a year's use. I use the default ext4 for the root and home partitions.

4.If you're setting up a system with multiple drives then the SSD is for the operating system, the programs("apps") and some of the most frequently accessed files. HDDs store big unchanging media files and such.

5.Probably not reliably, don't expect any such features to work if you need them. But they do have SMART capabilities like on an HDD and gsmartcontrol on linux can show you these SMART reports.

6.All modern operating systems will be about the same here. I would note that on Mint, I would avoid using swap partitions at all, and set the swappiness very very low indeed, the system will be slower and in some circumstances if your RAM is small there is a risk that something like loads of browser tabs open can freeze your PC, but avoiding swapping prolongs SSD life.

First run
sudo apt install --no-install-recommends smartmontools
and
sudo apt install gsmartcontrol

Do "sudo nano /etc/sysctl.conf" and add the line vm.swappinnes=15 at the very bottom . Then restart and confirm ths has worked by using "cat /proc/sys/vm/swappiness" . Check trim is set to be working: " cat /etc/cron.weekly/fstrim ".

Use "sudo smartd -i 108000" and also edit /etc/default/smartmontools to add the line "smartd_opts="--interval=40800"" at the very bottom. Check these have worked by using "systemctl status smartd.service". The idea behind these commands is not to overstrees the SSD by taking SMART readings too frequently. I thought I rememebred having to edit "/etc/smartd.conf" as well, but I checked  that file on my system and can't see any changes within in which look like I made them, so maybe nothing there is needed.

I also make some edits in Firefox and Chrome to make them use RAM for cache rather than using any SSD space for this, see here: https://easylinuxtipsproject.blogspot.com/p/ssd.html#ID9   https://easylinuxtipsproject.blogspot.com/p/ssd.html#ID10

If you run any virtual machines, take all the same actions inside them.

Never let the system enable hibernation, I think Mint is default against this anyway, Linux hibernation causes all manner of problems on many types of PC hardware well beyond just SSDs, but it is particularly bad for SSDs too.

All my notes in point 6 are for my Mint 21 system, there might be some minor changes when doing it for Mint 22.



As one further note, personally I chose a Western Digital "Red" grade 1TB SATA (my machine hasn't got more modern slots) SSD. It is a bit more professional grade than others and is technically designed to be a cache for a NAS. It does not have the weird "give up fast and get on with other things if there is a read error" mode that most NAS drives have, this drive should keep trying for a long time to re-read if it has any trouble as would any normal desktop drive. It has never had a firmware update made available for it, which suggest there were never any errors made when develeoping its firmware. Firmware updates for any SSD can be troublesome on Linux, so I chose one where I knew I would never need to.

There is one big problem with SSD comparison tables, manufacturers sometimes change the internal design of an SSD without changing the model name/number. The same brand of the same model of SSD might use different flash cells (different grades, different manufacturers...) and a different controller than the same model from months earlier.

Thanks, that's a lot of information difficult to understand for me.
I need time to read repeatedly.

[Greybeard]

Regarding ZFS or F2FS:

It seems to no longer be supported in Linux Mint 22.x, so further discussion here is pointless.

[Zenith]

You'll have no problems with ext4.

[paulca]

I have ZFS as the base FS on my virtualised servers.

The defacto features are fine, but it does come with considerable costs.

The backup disk pack of 3x2Tb metal spinners is a single ZFS with 1 disc redundancy.  However, when you first boot the box up, the discs will trash and trash and trash for nearly an hour.  Querying the ZFS subsystem it's doing things.  I can't remember if it's defragging or purging out of date blocks but it's long and slow.  Similarly writes to those discs are not exactly steeler speed either with a LOT of "chatter and clatter" when dumping large multi-giga byte "sequential files" which should be perfect for the write ahead cache, however I believe the way ZFS works, it doesn't care about your HD optimisations and expects you have a memory cache of insane size.  "Only enterprise needs apply" type thing again.  When used in less demanding situations it becomes a burden not a help.

That said, LVM doesn't come far enough and is a bit fiddly.  I looked at alternatives and was pointed back to ZFS for many reasons.

Going back to completely raw partitions... thats ugly too.

EDIT:  The other annoying things about it is.  ZFS is not a native linux filesystem.   ZFS has it's own file cache which does not intergrate with the Linux asynchronous filecache and so is not reported properly by memory stats.  The kernel has little control over the ZFS cache.  This means you basically need twice as much memory in the server.

[shapirus]

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.

Here's an example. KDE, two firefoxes (two separate profiles), Telegram, Viber, some minor stuff like syncthing and keepassxc. No special software, just a bare minimum of (atrociously bloated) modern desktop software. 32 GB is barely enough. Run something extra, like a game, a photo or video editor, a CAD program, and you'll need 64 GB or more, that is, if you want to minimize swap usage.

Code: [Select]

$ free -h
               total        used        free      shared  buff/cache   available
Mem:            62Gi        21Gi       8.0Gi       423Mi        34Gi        41Gi
Swap:             0B          0B          0B


[Greybeard]

Which one(s) would you buy?
https://geizhals.eu/?cat=hdssd&xf=7177_2.5%22~7185_3D-NAND%20TLC~7185_3D-NAND%20TLC%20(eTLC)~7185_2D-NAND%20TLC~7193_ab%20500TB~7525_SATA~8250_1TB%20(960-1024GB)&pagesize=100&offset=0&fcols=7176&sort=p&hloc=at&hloc=de&hloc=eu&hloc=pl&hloc=uk

Which manufacturers would you avoid?

Samsung and the big names are the safest choices, but they too put out some "value engineered" products, because there's always a market for cheap. The warranties are worth noting. I'd avoid anything DRAM-less or with QLC for serious use.

I thought so too, but Samsung is apparently discontinuing SSD production.
What happens in case of a warranty claim? Will I get a replacement from a third-party manufacturer, or do they have NOS stock?

https://www.heise.de/en/news/SSDs-and-eMMC-Samsung-apparently-discontinues-MLC-flash-production-10419217.html

https://www.hardwareluxx.de/index.php/news/hardware/festplatten/66260-mg-mlc-ist-nicht-gefragt-samsung-stellt-produktion.html

[paulca]

Code: [Select]

$ free -h
               total        used        free      shared  buff/cache   available
Mem:            62Gi        21Gi       8.0Gi       423Mi        34Gi        41Gi
Swap:             0B          0B          0B


Compare:

Code: [Select]

paul@UM560XT:~$ free -h
               total        used        free      shared  buff/cache   available
Mem:            13Gi       3.2Gi       7.1Gi       100Mi       3.3Gi       9.9Gi
Swap:          2.0Gi          0B       2.0Gi
paul@UM560XT:~$ vmstat
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 0  0      0 7423532 203856 3213276    0    0    27    17  219  365  2  1 97  0  0

"Ubuntu 22.04.5 LTS" on a MinisForum UM560XT.

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

The only reason there is a swap at all is because I used the auto installer.  I don't have any swap on anything I installed as I don't need it.  Never touch it and it causes issues with virtualisation.

EDIT:
Other examples...
A very busy dev server (VM) running KDE, though nobody has an open X session.

Code: [Select]

paul@linux-dev-vm:~$ free -h
               total        used        free      shared  buff/cache   available
Mem:            15Gi       1.9Gi        13Gi        19Mi       617Mi        13Gi
Swap:          2.0Gi          0B       2.0Gi
paul@linux-dev-vm:~$ uptime
 11:40:29 up 4 days,  8:25,  1 user,  load average: 0.27, 0.06, 0.02

A very busy, 100% allocated virtualiser host with swap disabled:

Code: [Select]

root@pve:~# free -h
               total        used        free      shared  buff/cache   available
Mem:            62Gi        59Gi       2.8Gi        66Mi       283Mi       2.5Gi
Swap:             0B          0B          0B

Why is it safe to run it that close to maxed out without swap?  About 50% of that "Used" is actually the dynamic ZFS cache I discussed prior.  When that is accounted for the host is only about 55% allocated.

[Greybeard]

In case I forgot to mention:
These are all older PCs with a SATA interface and 4-8GB of RAM, some already at the limit (laptops).

[paulca]

In case I forgot to mention:
These are all older PCs with a SATA interface and 4-8GB of RAM, some already at the limit (laptops).

The same "EVO" range are available as standard 2.5" SATA disks.  The interface is limited to 6Gbit/s IIRC, someone help if wrong.  Anyway, for spinning metal this is never a bottleneck, but for a good SSD on a SATAIII link it can be.  tldr; it will still be significantly faster (order of magnitude in real terms) to an old iron spinner.

Even with a swap partition you will be fine for most light to moderate use.