Does TRIM on SSD/NVMe hurt life of drive?

[VulkanBros]

Does TRIM on SSD/NVMe hurt life of drive?

 

[Wirko]

No. It should actually improve SSD life a little, especially if it's mostly full most of the time.

 

[SPDIF]

No, TRIM does not hurt the life of an SSD. TRIM is a command that tells the SSD which blocks of data are no longer in use and can be erased. This helps the SSD to perform better and last longer, as it avoids unnecessary write operations and frees up space for new data. On the contrary using an SSD without TRIM can hurt the life of the SSD.

 

[VulkanBros]

My thought also, but was on a IT Tech summit, and there was a guy from Lenovo Server Group, mentioning, that it was not all good, but maybe i didn't hear all of what he said. Maybe it was with drives in RAID, or some other drive system....

 

[SPDIF]

Most RAID (Redundant Array of Independent Disks) setups do not support Trim, that's right.
But they are working on that, to change this.

 

[VulkanBros]

Most RAID (Redundant Array of Independent Disks) setups do not support Trim, that's right.
But they are working on that, to change this.

Are you totally sure on that? On hardware level I think it is correct, but seen from the OS, it can't see it's RAID, and I am pretty sure, it can be defragged.
But maybe you are right, if the OS can't see it's a SSD/NVMe, then it is not possible to run a TRIM command...

 

[SPDIF]

yes, but de-frag is not same as TRIM. It is the RAID system itself that does not know the TRIM command. But there are now some that support the TRIM command.

TRIM does not work on all RAID setups, as some RAID controllers or software do not support or pass the TRIM command to the SSDs.

To check if your RAID setup supports TRIM, you can use tools like CrystalDiskInfo or SSDLife.

 

[Wirko]

The organisation of data in RAID 5 (distributed parity) is messy and details are not standardised. One sector (as the OS sees it) is mapped to several sectors on the disks, but at the same time several sectors (as the OS sees them) are mapped to one sector on the disks. I'm wondering if TRIM can work properly at all in hardware-based RAID 5.

 

[SPDIF]

If the raid system supports it, it should work i guess. But i did not tested that yet, so it's just guessing.

The only thing i know is that TRIM isn't a requirement for using RAID 5 with SSD. The storage controller is going to rewrite old block's, clearing cells out over time, in a manner that's agnostic to the storage technology. That's what my co-worker told me.

 

[unwind-protect]

ZFS with its zraid levels supports trim and should usually be better than hardware raid controllers anyway.

 

[SPDIF]

I know many people still use RAID, also in NAS systems, but sooner or later i think it will be all ZFS.

Both are different technologies that have their own advantages and disadvantages. RAID and ZFS are not mutually exclusive, as ZFS can use RAID levels within its pool of disks.
I also know that ZFS does not work good with hardware raid controller. it's interfere with its direct access to the disks. And raid is just more widely supported at the moment.
You also need a faster cpu and more ram with ZFS, because it is doing more operations with the data, and not all people are waiting on that to change all the hardware again, with is costly.

 

[Wirko]

If the raid system supports it, it should work i guess. But i did not tested that yet, so it's just guessing.

The only thing i know is that TRIM isn't a requirement for using RAID 5 with SSD. The storage controller is going to rewrite old block's, clearing cells out over time, in a manner that's agnostic to the storage technology. That's what my co-worker told me.

In some cases, trim can not work, even in theory. It can't work on parity data, which is either distributed over all disks (RAID 5) or stored on a dedicated disk (RAID 4). Of couse SSDs still work without trim, and SSDs like the WD Red SA500 or SN700 are better optimised for that situation, they have more overprovisioned space, I guess.

For parity you want an SSD with good performance and high endurance, since parity can never be trimmed, even when LT adds trim support for the array, also it will have to endure much more writes than the array devices, I would recommend a good NVMe device or an enterprise level SSD.

SSD Array for unraid?

forums.unraid.net

Unlike many RAID systems, SoftRAID supports TRIM commands on all SoftRAID volumes.

But that's mentioned only in the RAID 1 section, not in the RAID 4 and 5 section.

Raid levels - OWC Software Store

RAID 0 – Stripe SoftRAID stripe volumes give you the highest possible performance. However, they provide no protection from disk failure, regular backups are required for RAID volumes, because there is no redundancy. When a disk fails on a stripe volume, you will lose all the files on the...

softraid.com

Some SSD models do not support SSD TRIM for storage pools configured in RAID 5 and RAID 6. To find out if this applies to your SSD models, look them up on the Synology Products Compatibility List and check the Notes section for unsupported features.

Well, "Some". That's interesting.

Enable SSD TRIM | DSM - Synology Knowledge Center

Synology Knowledge Center offers comprehensive support, providing answers to frequently asked questions, troubleshooting steps, software tutorials, and all the technical documentation you may need.

kb.synology.com

SSD Trim is only performed on solid state drives that belong to a RAID 0, RAID 1, or RAID 10 group.

Storage Global Settings

 

[Pickles Von Brine]

Without trim your drive would die a hell of a lot sooner. TRIM allows cells not being used to be cleared. Now, when you go to write data your drive isnt moving data that was deleted but still there to write things.

I know many people still use RAID, also in NAS systems, but sooner or later i think it will be all ZFS.

Both are different technologies that have their own advantages and disadvantages. RAID and ZFS are not mutually exclusive, as ZFS can use RAID levels within its pool of disks.
I also know that ZFS does not work good with hardware raid controller. it's interfere with its direct access to the disks. And raid is just more widely supported at the moment.
You also need a faster cpu and more ram with ZFS, because it is doing more operations with the data, and not all people are waiting on that to change all the hardware again, with is costly.

Oh and please tell people to not backup their stuff. I still need a job XD Peoples' mistakes turn into my cake. (sees your sig)

 

[:D:D]

FWIW fitted 2 mSATA cards in iRST RAID0 which supports trim. Unless things have changed then TRIM is not supported in RAID 1. Anyway, fitted about 10 years ago although less than 6 years of on time I think, as hours on time seems to have rolled over but available on the HDD. Pretty much on 24/7 over the last 2 years as measuring some sensors every 6 and 10 minutes. Been rock solid so far.

 

[chrcoluk]

Trim tells the drive ahead of time which blocks are no longer used, otherwise it only finds out when the file system tells it to overwrite those blocks.

So the effect is the drive can wipe ahead of time mitigating any performance impact of an erase during a write, and this should also allow it to have better wear levelling which if anything improves life of drive.

Scenario, drive was 100% full, has factory supplied 3% over provision, so 3% of capacity used for wear levelling (very small), you then wipe 20%, without trim, there is still only 3% available as drive doesnt know the data can be erased, with trim, it erases the 20%, and now has 23% available for wear levelling.

 

[Wirko]

Trim tells the drive ahead of time which blocks are no longer used, otherwise it only finds out when the file system tells it to overwrite those blocks.

So the effect is the drive can wipe ahead of time mitigating any performance impact of an erase during a write, and this should also allow it to have better wear levelling which if anything improves life of drive.

Scenario, drive was 100% full, has factory supplied 3% over provision, so 3% of capacity used for wear levelling (very small), you then wipe 20%, without trim, there is still only 3% available as drive doesnt know the data can be erased, with trim, it erases the 20%, and now has 23% available for wear levelling.

That's a very good brief explanation, let me just add that more overprovisioned space should also result in lower write amplification because the controller never, or rarely, has to erase partially written blocks to make space for writing.

 

[:D:D]

The RAID0 I used spanned 85% of the drives with the other 15% left unallocated for the controller's use. Trim works on a sector basis, typically 512Bytes while the drive uses pages to read and write and blocks contain a number of pages. So trimming may leave partial data on blocks that is still being used and with erasing done at block level will be up to the drive firmware when it decides to shuffle stuff around if need be to make free blocks while trying to keep erasures even across the device and to a minimum.

BTW that 15% wasn't calculated but a compromise between usable space for me and giving the controller a bit more to work with. Some people go much higher with provisioning.

 

[NixonP]

That's a very good brief explanation, let me just add that more overprovisioned space should also result in lower write amplification because the controller never, or rarely, has to erase partially written blocks to make space for writing.

One reason trim is important and what confuses many everywhere and in this thread is an SSD cannot rewrite to a written sector on the fly like a spinner does, i.e. if literally just 1 1 and 511 0's are in a sector that sector must be fully tosssed into cache to do anything with it (this is also why read-wear leveling becomes important), read, combine, delete, anything that modifies data it requires it be written to a different sector, CANNOT just change a 0 or 1 anywhere, once a sector is written the only thing that can change that sector is the TRIM command which wipes the entire sector and lets the controller know that sector is free for writing again. So you open a txt document that is illogically 512 bits, change a single letter and save, that 512bits txt file is written to a new sector and the controller or the OS will mark the old sector, "to be trimmed" and trim it as scheduled or immediately if you foolishly have no free sectors (you just burned part of your drive if it is in use and you filled it to capacity with 0 overprovisioning) ssds are rated in writes to sectors despite mtbf ratings, usually any current ssd can do 400-700 rewrites for any sector (there will always be a trim between rewrites) I ruined my first ssd cause I filled it up without any overprovisioning while also havinging page on it and that is simply stupid and is the best way to get ssd sector life into the seconds zone.

 

[:D:D]

I think usually when a sector is marked with trim then any reference to it is returned as zero's or ones depending on firmware, from cache and not read from flash media, nor do zero's need to be written to trimmed flagged media.