How reliable are SSDs in Linux systems?

[Greybeard]

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?

[BitBangBro]

+

• Rewritability depends greatly on the filesystems chosen. With SSDs nowadays it makes no sens to use anything but F2FS, whis his CoW filesystem, tailored to fit NAND FLASH specifiically.
  Failing that, if you insist on using traditional fileystems, use XFS ( but check the mount options first). Another good option might be BcacheFS, if you want to use two-tiered mount, with SSDs acting as a cache to a much bigger HDD group, possibly bound into a raid before the BcacheFS initialization (mdadm or HW RAID).
• If/when you can, use mount options like "lazytime" or even better "noatime", so that the access time doesn't get written to the SSD every time you read the file.
• When partitioning the SSD, leave some part of it free - unpartitioned space. Controller will use it as a storage for bad pages relocations. This is most of the the trick that separates write-intensive professional media (U 2/3 etc) from cheap consumer media. They both use the same underying storage media.
• If you are using something that intensely writes to the media, it makes sense to use separate tmpfs (RAM based) mounts for this. That is, if you have enough RAM. typical use  case for this is on Gentoo for temporary /var/tmp/portage map, where packages get unpacked, compiled etc and then deleted after inatallation. Another examples would be on most distros /run, /dev/shm, /tmp, /var/run, /var/tmp etc maps
• 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.

[Karel]

I have been using SSD's for more than 10 years with Linux and they work just fine.
I always use ext4 filesystem with "defaults" option.
However, I do use Samsung Pro SSD's because I don't want to risk my data.
But you should make (incremental) backups of important data and documents to an external backup drive/NAS/cloud anyway...

[BitBangBro]

I have been using SSD's for more than 10 years with Linux and they work just fine.
I always use ext4 filesystem with "defaults" option.

That's fine as long as you don't approach any of the critical limits.
As long as controller can do its brainfuck schemes to spread the writes and there is no significant write amplification, who cares.
Especially if you plan to obsolete your gear much earlier than it is bound to hit any of its longevity limits.

For anyone else, F2FS is the king.

[Greybeard]

1. Rewritability depends greatly on the filesystems chosen. With SSDs nowadays it makes no sens to use anything but F2FS, whis his CoW filesystem, tailored to fit NAND FLASH specifiically.

I understand.

Failing that, if you insist on using traditional fileystems, use XFS ( but check the mount options first). Another good option might be BcacheFS, if you want to use two-tiered mount, with SSDs acting as a cache to a much bigger HDD group, possibly bound into a raid before the BcacheFS initialization (mdadm or HW RAID).

I still try to understand...

2. If/when you can, use mount options like "lazytime" or even better "noatime", so that the access time doesn't get written to the SSD every time you read the file.

I understand:
https://wiki.archlinux.de/title/Noatime

3. When partitioning the SSD, leave some part of it free - unpartitioned space. Controller will use it as a storage for bad pages relocations.

I understand.
Is this done automatically by the controller, without requiring any configuration by the user?
What percentage would you leave free?

This is most of the the trick that separates write-intensive professional media (U 2/3 etc) from cheap consumer media. They both use the same underying storage media.

What does U 2/3 mean?

4.If you are using something that intensely writes to the media, it makes sense to use separate tmpfs (RAM based) mounts for this. That is, if you have enough RAM. typical use  case for this is on Gentoo for temporary /var/tmp/portage map, where packages get unpacked, compiled etc and then deleted after inatallation. Another examples would be on most distros /run, /dev/shm, /tmp, /var/run, /var/tmp etc maps.tmpfs

No, RAM is low because the machines are old (one is very old).

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

How can I reliably identify them?

[Greybeard]

For anyone else, F2FS is the king.

Is it possible for a Linux beginner to do this?

[thephil]

• Consumer grade SSD offer less lifespan than old style harddrives. In practice, it probably does not matter as you will most likely not use a laptop or desktop system for too many years. Regular backups are mandatory, as they were with hdds. In contrast to hdds, ssd tend to lose data when left unpowered for years. The default filesystem is fine for normal use. Servers are a different story, as the typically run under much heavier load and the work of lots of people rely on them. Accordingly, one would use more expensive drives in RAID systems or something like ZFS in those scenarios. For desktop systems it's not worth the hassle in most cases.
• Depends on what is important to you. Read/write speed, capacity, MTB come to mind.
• The size is typically not limited by the OS but by the user data you want to store. Also system storage requirements highly depend on the kind and number of programs you install. My current Debian Linux installation on my Laptop uses a total of 676GB storage. 641 GB of that are user data.
• Yes. You could try moving paths like /bin, /usr/bin etc. to another disk, but I see no advantage in doing that on a desktop sysem.
• Just like hdds, ssd typically support SMART data which can be checked using the usual tools (e.g. smartctl.)
• The specific LINUX distribution does not matter much, as all of them use the LINUX kernel and thus work pretty much the same under the hood and can be configured the same way. Personally, I don't make SSD specific changes to my config and have yet to suffer an SSD failure. There are pages on the interweb that discuss potential optimizations like reducing certain write operations or even running some disks read-only. Given the price, I don't bother and would be happy to get a new drive and restore from backup when a failure occurs. (A brand new Samsung 1TB  EVO pro 990 is about 124€ here – compare that to your hourly rate...)

[BitBangBro]

Failing that, if you insist on using traditional fileystems, use XFS ( but check the mount options first). Another good option might be BcacheFS, if you want to use two-tiered mount, with SSDs acting as a cache to a much bigger HDD group, possibly bound into a raid before the BcacheFS initialization (mdadm or HW RAID).

I still try to understand...

Linux ain't Windows. It has many options for chosen filesystem. Some are more viable and tested than others.
EXT4 ain't the only game in town.

XFS is traditional "old-timer"/oldie-but-goldie that has been brought in by IBM and has been with Linux from its early days. It has grown into kernel and can use its many special tricks.
Like exporting whole disk as a block over Ethernet and many other low-lewel-magic stuff that give it edge for special uses.
As a rule, it's faster than EXT4 for working with big, heavy files, but less nimble with bazilin of small files etcetc.

There are others. BcacheFS is a new one that is expected to seriosly compete with ZFS for people with advanced needs.
One of the tricks that it brings is to combine HDDs with SDDs into one or more virtual drives, which effectively uses SSDs (being fast) asa a cache into a pool of much bigger (but slower) HDDs.
But there are others. LIke redundancy ( RAID in the pool - your data survives the death of any disk in the group), snapshotting ( ability to "freeze" the fileystem's look at any poitn in time, even though someone might have been writing after that) etc etc.

But if your needs are simple (like most) and don't need any of that "datacenter stuff", F2FS might be your choice for SSD.
Other choice might be (very similar) NILFS, but that hasn't been updated lately and not that many people are using it nowadays.
IMO, for SSDs it's by far best to stick wtih F2FS for 99,99% people using SSD nowadays.

What does U 2/3 mean?

It's a form-factor of proefessional SSDs that connect directly to PCIe lanes.
It differs form consumer M.2 stuff (besides fom-facotr) in that it can endure MUCH more writes. Less sturdy ones are usually declared for 1DWPD ( one disk write per day) and beefier, wirte intensive models can do 3DWPD.

How can I reliably identify them?

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.

[BitBangBro]

For anyone else, F2FS is the king.

Is it possible for a Linux beginner to do this?

Ofcourse. There is not THAT much to this.
You just need kernel support for this (most distros have it and much more), userland tools for un/mounting, "formatting"/initialization, file check ts ( f2fs-tools) and the right moutn option in /etc/fstab.
And you have to initialize it onto a partition with something like

mkfs.f2fs -l <MY_LABEL> /dev/<MY_DEVICE_AND_PARTITION>

[kripton2035]

you can use the best drive model here, hdd or ssd, brand name or not, if you don't have backup(s) you will get screwed one day...
I've installed my first ssd in 2010, yes it was out of price at this time, but I never got back to hdd after that
for me and my many customers.
where have you been last 15 years to never have use an ssd !?!?

[BitBangBro]

I've installed my first ssd in 2010, yes it was out of price at this time, but I never got back to hdd after that

That said, HDDs still have PLENTY of good use.

Killer combo ishaving SSDs, caching MUCH larger HDDs bunch.
So you get many tens or hundreds  TB of space effectively mostly behaving closely to being one giant SSD.

HDDs have much more write cycles (practically infinite) and way better cold storage lifetime and reliability.

[Greybeard]

Where have you been last 15 years to never have use an ssd !?!?

I stopped building PCs as a hobby and maintaining them for my friends about 12-15 years ago because it became increasingly easy to set up systems with WinXP/7/10, and Microsoft still managed to spoil the fun for me of delving as deeply into the system as I did back in the days of Win95/98. All that mattered to me was that the system was usable; I no longer cared why.

[JohanH]

Haven't had an SSD die since my first SSD, an OCZ drive some time around 2010 (they had some issues). And I've been using Linux since SSDs didn't exist yet.

Not related to Linux (although I used it with Linux), I just had a 4 year old Samsung PM981a 256 GB (OEM) die. It was not flash/memory issues. It started by being absent at startup at times, so the PC had to be switched on and off. Now it can't be found any more. I'm suspecting hardware issues, because the PCB looks bent. Fortunately, it was a spare one.

[Simmed]

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?

[wilfred]

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?

I use Linux Mint and I just did what most beginners do. I created a LM boot USB and pressed install with a SSD in the machine. It doesn't really matter how long they last if you aren't prepared with backups. They last longer if you have backups. 

Don't overthink it.

I suggest a separate HDD for making a backup onto. I use Timeshift. But when I do a significant upgrade I copy the SSD to a spare one and swap them and upgrade onto the spare. It can save time if you have to climb out of a hole of your own creation.

I also suggest partitioning during LM install to have a root file system and a separate partition for your home partition. But you don't have to.

[thephil]

They last longer if you have backups. 

Don't overthink it.

 

Discs can smell fear.

[themadhippy]

I also suggest partitioning during LM install to have a root file system and a separate partition for your home partition

Id go further and suggest a separate  drive for home.My main machine is getting on for 10 years old and is  still on the same ssd's it was built with.

[Zenith]

I've been using SSDs with Linux for ten years and I've never had any problems with them. I haven't come close to the TBW limit on any of them. None of them have failed or thrown up errors. What's happened is that after four or five years, the drive has become full, and higher capacity drives were available for a reasonable cost, so the SSD was retired and used for something like playing with thin clients.

Unless you are also using an HDD which can host swap, you need to make sure the PC has enough RAM to avoid swapping, because that can make a huge amount of writes to the SSD. 16 GB seems enough. 8GB is marginal, IMHO, if you have Firefox with a lot of tabs open and leave it for a long time.

Of course, you need to make regular backups, as with HDDs.

[Greybeard]

Perhaps I should have mentioned that I've had a lot of defective USB thumb drives and SD cards over the past few years. Most of them were no-name brands, but also some well-known ones.
I've also read here in the forum about the many flash failures of the HP/Agilent 34461/34465 DMMs, which hasn't exactly increased my confidence in flash memory technology.

[thephil]

There is a reason, thumb drives are substantially cheaper than SSDs of the same size.

[David Hess]

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)?

It is possible to wear out a smaller SSD with excessive writes within a year, but generally normal workloads will allow them to last 5+ years.  Some of my small SSDs are approaching wearout at 5 years.

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

Besides power loss protection, I check the endurance which is typically listed as GBW or TBW (gigabytes or terabytes written), and I calculate a figure of merit which is how many times the SSD can be overwritten.

So for instance the Crucial MX500 series can be overwritten about 350 times, while the Crucial T500, T700, and T705 can be overwritten 600 times.  Some drives are much worse.

Keeping mind the above, larger SSDs can accept more total bytes written than smaller ones, so besides capacity, a larger SSD will last longer.  If you apply lots of write traffic to an SSD, even when only using a fraction of its total space, and larger capacity SSD will last longer than a smaller capacity SSD.

[Greybeard]

(Attachment Link)

The WD SSD marked in red is shockingly unreliable compared to the others.
https://www.backblaze.com/blog/ssd-edition-2023-mid-year-drive-stats-review/

[Zenith]

There is a reason, thumb drives are substantially cheaper than SSDs of the same size.

Yes and thumb drives and SD cards were designed with a particular use pattern in mind, which does not include lots of random reads, writes and erasures. SSDs have quite elaborate wear leveling and built in over-provisioning. Thumb drives use a random allocation of spare blocks, or so I've read.

I have a number of SD cards, used in cameras and Raspberry Pi, and thumb drives used to transfer data between systems and as boot drives, and I've yet to have one fail. Most of those are over five years old. They are mainly Kinsgston or Toshiba. I believe there are up-market SD cards for dashcams and other intensive applications.

[Greybeard]

For anyone else, F2FS is the king.

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

[Zucca]

I dare to say good SSD from reputable companies are today vastly more reliable than old spinning disks.
Buy good SSDs, copy all your data from old and slow spinning disks into them and enjoy your new faster life without Microsoft.

F2FS or whatever is not a must do not waste your time on that. The default EXT4 (or the my favorite more advanced ZFS, but you need more RAM) is perfectly fine.

Backup ARE a must, take your time and (let the system, remember you are in Linux now the option are endless...) do backup regularly.