Originally posted by treba
View Post
Announcement
Collapse
No announcement yet.
Fedora 33 To Be Released Next Week
Collapse
X
-
Ssd and nvme has a tendency to do bit flips and such. Btrfs csum is invaluable in those cases! Doesn't matter if it is on laptops or nas boxes or desktops. I had a corrupt block on my Samsung SSD the other day. Btrfs saved my ass there because it happened on my boot/vmlinuz
Didn't Michael post some benchmarks recently that shows btrfs is not slower than ext4 is majority of tasks? Not that it really matters that much these days.
Regarding sqlite. Enable WAL and it is nice on cow filesystems! https://sqlite.org/wal.html
Someone mentioned LVM for csum/integrity. That is _abysmally_ slow! Also still buggy.
Last edited by S.Pam; 23 October 2020, 06:33 AM.
- Likes 1
Comment
-
Originally posted by flower View PostI still see no reason to use btrfs and will stick to ext4
Performance? Well ext4 is better in nearly task
Fragmentation? Ext4 wins by a big and performance noticeable margin (eg firefox sqlite db)
Snapshots? Lvm
Dedup? Lvm
Raid? Lvm
Bitrot detection? Lvm (no corrections though, but well i have backups)
Compression? Btrfs is a winner. But i have enough space and my biggest files are not compressable anyway
Did i miss something? Or is btrfs just a solution without a problem
LVM snapshots are fail-deadly and require manual space management.
LVM raid is even worse than mdraid. It has lots of missing features, and, like mdraid, only protects from full disk outages, not bitrot.
LVM deduplication is experimental and... well, that's it, also snapshots caveats apply.
LVM bitrot detection — ??? Does it even exist?
By the way, even if you pair mdraid/lvmraid with dm-integrity to handle bitrot, its answer to any form of corruption is "let's throw away the whole disk and rebuild the array from scratch", which is not just counterproductive, but also dangerous. Considering modern disk sizes, it is statistically non-negligible that you will experience another read error while rebuilding, throwing the whole array into the trash bin.
So yeah, you absolutely did miss something. LVM is a technology with lots of deficiencies that covers only half of usecases.
Therefore, rather than the populist lie of "LVM does everything you ever need, btrfs is a solution without a problem", the actual problem is pragmatic — there are two technologies, each with its benefits and deficiencies. Given two competing technologies where each requires work, it makes sense to focus on stabilizing the modern technology rather than endlessly beating the dead horse.
- Likes 8
Comment
-
Originally posted by horizonbrave View Post
It's a filesystem not sex with your girlfriend/boyfriend.. all this obsession for performance it's non-sense, unless we're talking about an abysmal difference!
Whatever turns your libido man
I'd rather have tons of features because I don't notice the, pulling numbers, 5-10% difference in speed since spinners are slow already and SSDs are usually fast regardless. My spinning mirror feels fast enough for games and my SSD is plenty fast for a desktop. Worst case scenario, I'll have to add backing SSD cache to the mirror because some game doesn't load fast enough. I like knowing advanced features like that are an option more than I want a slightly faster file system.
Plus once encryption and/or compression are factors they usually have more to say in the r/rw speed than the actual file system being used...every file system will slow down if either of those are set high enough. So, again, I figure if I'm trying to save stuff securely or compactly, might as well pick a safer file system over a faster file system. With the goals at hand the speed differences are more than likely negligible.
- Likes 1
Comment
-
Originally posted by skeevy420 View PostOr why I use ZFS these days. Y'all feel free to replace that with your file system of choice because I'm not trying to start a file system argument.
- Likes 1
Comment
-
Originally posted by Spam View PostRegarding sqlite. Enable WAL and it is nice on cow filesystems! https://sqlite.org/wal.html
Comment
-
I know this is tilting against a windmill, but: this is an exactly on time release, and the "Target #1" date is the one you should plan on. We would prefer to have it ready a week early (which is why that's called a preferred date!), but not being early is not a failure.
Comment
-
Originally posted by cen1 View PostFedora is getting their releases out so frequently and on time now that it stopped being exciting to upgrade. I say that in the most positive way.
Comment
Comment