You are not logged in.
Hi,
This article mentions SSDs zero-ing out data due to TRIM:
https://blog.algolia.com/when-solid-sta … hat-solid/
And the SSD which I happen to use, Samsung SSD 850, is mentioned in the list.
"Samsung SSD 8*" is also mentioned here: https://github.com/torvalds/linux/blob/ … 4109-L4286
The article seems not specific enough, does anyone know how to interpret this:
Is my data gonna be destroyed by this SSD in ArchLinux? Should I throw the disk away and get a different one, or is Linux able to use this disk and not zero-out my data?
Thanks!
Last edited by aardwolf (2015-06-16 10:31:20)
Offline
TRIM discards data, that's what it does. Even if it does the right thing(tm), it prevents you to get deleted files back. If you delete something on an SSD (and your filesystem uses the discard flag), it's really gone. Is it worth it just for seeing a few more points in some benchmark? For 24/7 hot database servers it might be (as long as it works reliably). At home it usually isn't.
Is my data gonna be destroyed by this SSD in ArchLinux?
Probably not, but like any other storage media it can die completely at any time, so you should have good backups in any case.
Offline
Yes I did use the discard flag, so maybe I shouldn't, but that'd affect performance right... ![]()
I guess I just need to understand the difference between this queued and unqueued TRIM better.
Anyway, thanks for your reply ![]()
Last edited by aardwolf (2015-06-16 13:16:58)
Offline
Disabling TRIM also shortens the life of the drive. Having a ready supply of empty blocks keeps the write amplification down.
Offline
Disabling TRIM also shortens the life of the drive.
But what's the life of the drive?
http://techreport.com/review/27909/the- … all-dead/4
Over the past 18 months, we've watched modern SSDs easily write far more data than most consumers will ever need. Errors didn't strike the Samsung 840 Series until after 300TB of writes
My SSD is currently ever-so-slowly reaching the 4TB mark, after 10k Power-On-Hours. If it goes on like this it will fail in >80 years. By that time it'll be ancient technology (it's ancient technology today since it's only 64GB, which was the affordable size some years ago) and I'll be dead anyhow. Write amplification would not change the picture much.
At the same time with TRIM you sometimes risk the life of your data. You can always buy a new drive but not new data if your backup turns out to be bad as well.
I use TRIM myself (monthly fstrim, NCQ disabled since my SSD and HDD perform better with/out it and most bugs seem to be NCQ related in some way). But at the same time I think that TRIM is vastly overrated due to overmuch worry about SSD lifetimes.
In Linux *everything* trims, even without asking you for consent. You make a new filesystem or swap and its implied you want the entire disk trimmed. Mistakes on SSD are more deadly than they used to be on HDDs.
Offline
My SSD is currently ever-so-slowly reaching the 4TB mark, after 10k Power-On-Hours. If it goes on like this ...
I completely agree with your overall point. But estimating future data use on current data use is almost always bound to be pointless. Not too many years ago one might not be able to imagine ever needing a drive that could even store more than a GiB - now one can easily download that much just to get a little music.
But just the same - by the time you reach the write limits of a modern SSD, times will have changed so much that you'll either be looking at buying the newer bio-pack-memory-cell drive, or a quantum-photon-encasement drive, or some other thing we couldn't yet imagine. Or at the very least, you want to trade in your SSD for a larger capacity one - if your daily web-use is on the TB scale, you'll probably need a bigger drive to store whatever files you are working with.
So while we can't predict the future - we can be pretty sure that current SSD obselescence will not come from reaching their read/write limits.
EDIT: for reference, I have two machines that run on SSD only. I backup regularly, but no more so than I would with an HDD.
"UNIX is simple and coherent" - Dennis Ritchie; "GNU's Not Unix" - Richard Stallman
Offline
I only have a 64GB SSD, and the clunky stuff lives on HDDs. Very long lived HDDs at that, old WD Greens 2TB bought before the floods that made prices skyrocket I don't even remember how many years ago.
Sometimes I wonder if those HDDs are the last ones I ever bought, if SSD keep dropping prices like they did last year we might reach that point eventually.
SSD lifetime (due to writes) probably won't be an issue ever, the larger their capacity the more wear leveling it can do to extend life. It's more likely to die of a random defect...
Offline
@frostschutz
The problem with SSDs is data retention, if you leave them powered off for too long they will start to lose data. I suppose traditional hard disks aren't immune to this but the time scales are completely different.
@aardwolf
It should be safe to enable trim as the problem was identified and there is a workaround in place. It will also help the disk with its garbage collection, not having to move around data you don't need anymore helps to keep the number of P/E cycles down. What might be more important to you is that you may see slightly better performance for write intensive loads because the SSD will have a larger percentage of flash blocks ready to be written to.
R00KIE
Tm90aGluZyB0byBzZWUgaGVyZSwgbW92ZSBhbG9uZy4K
Offline