This paper shows btrfs having almost x16 more write amplification than ext4 or xfs.>https://ar5iv.labs.arxiv.org/html/1707.08514
>>109580997
i use xfs everywheredid i win did i win??
>>109581068xfs has to be top tier if redhat still defaults to it on rhel. Even though fedora has been on btrfs since before rhel 9, both rhel 9 and 10 default to xfs.
ssds have cache/hmb to buffer out the writes.
>>109580997Would certainly explain a lot.
>>109581164That would work regardless of file systems. It will certainly help increase the SSD's lifetime, but it doesn't change the fact that btrfs 'causes more physical writes in the longrun when compared to xfs.
>>109580997yeah no shit, you have literally 0 reason to use anything else that ext4 on your fucking desktop and server.
>>109580997more like BTFOfs amirite
>>109581357And how I do btrfs scrubs in raid1 with ext4? Also ext4 doesn't have transparent compression and snapshots. Useless ancient filesystem for boomers.
>>109581443>And how I do btrfs scrubs in raid1 with ext4?just don't do taht worthless shit>Also ext4 doesn't have transparent compression and snapshots.good. use tar.gz and dd>Useless ancient filesystem for boomers.they don't make them like they used to, modern filesystems are utterdogshit answers looking for problems to solve
>>109581443Snapshotting was never an issue.What btrfs does is impressive but it's not like you can't snapshot other filesystems with rsync.Drive space is the cheapest digital commodity and with ext4 and xfs you get longevity as well.What's the point of btrfs snapshotting if the drive will end up dying as a whole sooner anyway?
>>109581744>What btrfs does is impressivecrashing and corrupting data while ignoring bug reports?reminder that Tso is a literal subhuman, btrtfs will never ever be good>Drive space is the cheapest digital commodity and with ext4 and xfs you get longevity as well.fs compression is the most retarded idea you could ever add to a filesystem, it saves fucking nothing for extra power, absolutely worthless.>What's the point of btrfs snapshotting if the drive will end up dying as a whole sooner anyway?exa, don't use this shit
>btrfs is bloated cowshitIt's almost like the only reason anyone actually uses it despite being objectively inferior to everything before it is because it's being shilled hard by corporations like Red Hat, just like systemd and wayland...
I recently converted a drive to XFS from BTRFS for this very reason. It serves as a "semi-volatile" data drive for my home server (a bunch of containers and other shit running off of it). I happened to look at the SMART overview in proxmox one day and almost shit myself when I saw the wearout. Disregard the power on hours, the disk has been on 24/7 for a little under two years. Either way, the bytes written is absolutely fucking stupid. I'll keep using BTRFS for things that aren't used as frequently (couch gaming PC, etc), but for anything that will see even remotely frequently writes, fuck that. Pic related.
only had a fucked up FS with ext4 and btrfsbtrfs was repairable, etx4 was not. xfs ive never had problems with. also on desktop, ive never had a drive wear out. ive got ssds that are over a decade old and theyre still prob over 95% usable
>>109581744>it's not like you can't snapshot other filesystems with rsync.How in the fuck do you make an atomic snapshot with rsync? Rsync is a copying tool not a snapshotting tool
>>109580997>CoW filesystem has write amplification Wow, no way.
>>109581744>nooo, btrfs kills your disks>Drive space is the cheapest digital commodityPick one
btrfs is actually way faster on SMR disks fyi.
ext4, just like NTFS, wins by doing nothing, again.
Ok, tell me an alternative with the same featureset that doesn't have these issues.
>>109584077BTRFS with a bad update>boot into healthy snapshot taken right before the update >rollback>fixedEverything else>that filesystem backup you took doesn't actually work>that image you created months ago is out of date>your automated backup solution stopped working months ago>you're manually restoring a backup writing over shit you hope fixes it>you're actually just completely fucked
>>109584106My thing (good)>simplified positive scenarioYour thing (bad)>a very negative scenario>purposefully complicated>every thing that could go wrong went wrong>it's really really bad and your thing is bad and you should feel bad
>>109580997Jews
>>109581085Complete fucking retard here, but the idea that "business class" has roll-over to consumer grade products is a terrible premise to take. The needs are different, a business won't hesitate to top dumb 50k worth of SSD's and put in 50k more, if the cost can be justified through a tangential (quite literally, non-related) means, i.e. end of financial year spending, tax, charity etc.So to say, RHEL uses x, x is a business class, therefore x is better than y is fallacious.
make 50GB root partition 7GB install after 16x snapshots newb thinks ssd gone bad
Thanks, but I'm sticking to ext4
>>109580997UNUSED WRITES ARE WASTED WRITES
>>109580997is it possible to convert a boot drive from btrfs to ext4? or is this just another nothing burger.
>>109580997Yeah no fucking shit, btrfs is copy on write and ext4 and xfs aren't. btrfs is shit compared to ZFS but both have significantly higher writes than non CoW systems because CoW systems can't in place modify data and always rewrite the chunks fully before deleting them.>>109584079ZFS does most of the same shit better and is a proper battle tested enterprise file system. The problem is it's out of kernel and is a pain to set up on root.
>>109581744>it's not like you can't snapshot other filesystems with rsync. you can't retard. rsyncs doesn't do atomic snapshots. it just copies data.
>>109580997why the fuck are you posting a paper from 2017? there's also more than enough methods to reduce unnecessary write amplifications, as long as you know what you are doing.
>>109580997Thank you for the information. I was considering switching my system to btrfs, but was hesitating due to the weeks of customization required. I'll stick to ext4 and clonezilla backups instead.
>>109583999If you kill your disk you need an actual backup, and if you keep actual backups, you don't use btrfs because you're not a braindead zoomie.
Doesn't btrfs have ssd optimizations?
>>109585184How do you know that you are not backing up corrupted files without a proper filesystem that logs checksums? I run a btrfs scrub that verifies all the data checksums and then I take a backup to another disk. How do you verify that the data you are backing up is not corrupted? Even if you rsync backup with checksums you don't know if a file was corrupted 1 week ago because the checksums are new, they aren't logged checksums that happened when the data was written on the disk. Also ext4 doesn't have compression, that makes file reading and transferring faster and you have less free space too.
>>109585258>corruptiondont care. I don't use rsync. I image my drive and convert it to a compressed qcow2.>compressionwaste of cpu, anything that takes up enough space for me to be concerned about goes on external media. If I need compression there, I'll use it.
>>109580997>wrote 11Tb in a yearluckily it's my job-provided machine
>>109581068>XFSIf you want to lose your data when theres a random crash or outage, then you chose the right file system.
>>109581744>actually, btrfs kills your disk with writes>actually, having shit ton of writes by rsyncing isn't a problemat least choose a narrative and stick to it, you retarded RedHat goblin
>>109585184>and if you keep actual backups, you don't use [only usable filesystem that does checksumming], you instead use [filesystem that doesn't do checksumming]
>>109584979Honestly, what are the advantages of ZFS over btrfs, other than proper striped RAID support?On the converse, I'm pretty sure btrfs offers a lot more flexibility for editing live filesystems, adding/removing drives from a RAID/pool, shrinking or resizing, editing the mirroring settings etc., where ZFS for many things requires taking the pool offline and rebuilding it as far as I'm aware. Correct me if I'm wrong on this, but imo that makes btrfs better for home/desktop usage where you might change disks around for various reasons or might want to edit your setup, as opposed to e.g. a NAS you set up with a pre-planned set of drives and don't plan on touching for several years.
>>109585283>waste of cpuUnused CPU is wasted CPU, if you're waiting on I/O then compression could actually speed things upObviously this is very workload-dependent but it's also usually not harmful unless you're doing the opposite kind of specific workflow, where you're highly CPU-bound while writing large I/O in the background that isn't bottlenecking you. In my experience compression mostly harms latency-sensitive workflows but btrfs latency is already in the gutter compared to other filesystems so turning off compression is barely a fix at all, for those (which are relatively rare on a desktop) you really need a separate ext4 partition.
>>109581037is f2fs stable or will it corrupt my shit up?
>>109580997>Jayashree Mohan1 Rohan Kadekodi2 Vijay Chidambaram1im not trusting a paper by 3 pajeets.
>>109581816cope
>>109585588What the fuck is this chartHow does filesystem throughput depend on whether you're "Gamer", "Workstation", "Server" or "NAS"?
>>109585602it depends on what tasks you are doing, retard.https://www.youtube.com/watch?v=skqgEQrDpUA
>>109585571it's made by koreans and cheats on fsync
>>109585445This, XFS is only fast because it's a thin layer on top of>just write shit wherever dude idgaf
so what's the score if i disable CoW on btrfs?
>>109580997snapshots
>>109580997What do you expect from a filesystem whose name sounds like having a really bad time in the shitter?
>>109585778You probably need to disable both CoW and checksums, at least.A lot of the write amplification comes from the fact that SSDs amplify all writes, so if you write 1KiB of new data and then 32 bytes of updated checksum stored in a different location in the FS metadata, it might double the SSD wear since two sectors need to be touched.
