Linux File-System Benchmarks On The Intel Optane 900P SSD

[leipero]

niner I don't know, I don't use snapshots, actually, this is first time I did hear about that concept, and after reading for a bout 5-10 minutes about it I failed to see any benefit over traditional data backup (keeping current form, non-compressed or tool-only formats) for Desktop users, it just doesn't make any sense. As for rolling back updates, package manager? If incremental backup means adding/synching periodically in consistent or non-consistent manner, there are trillion ways of doing so, from scheduling background tasks to doing it at shutdown time. I can't answer last question, fsck should take care of that to my knowledge.

[caligula]

I don't know, I don't use snapshots, actually, this is first time I did hear about that concept, and after reading for a bout 5-10 minutes about it I failed to see any benefit .. it just doesn't make any sense .. package manager? .. I can't answer last question .. to my knowledge

You know, if you already publicly acknowledge that you're not qualified to answer, why are you posting instead of educating yourself.

[niner]

niner I don't know, I don't use snapshots, actually, this is first time I did hear about that concept, and after reading for a bout 5-10 minutes about it I failed to see any benefit over traditional data backup (keeping current form, non-compressed or tool-only formats) for Desktop users,

There is a huge difference between making a snapshot and running some incremental backup. The former takes about 0.2 seconds, the latter would take minutes at best (little change, fast backup system on fast storage), hours at worst. Accessing files in a snapshot is the same as accessing files in the main file system - they're just there. Just use cp and be done. Due to the near instantaneous availability of snapshots tools like zypper with snapper integration can automatically make a snapshot before every package installation and update. Something goes wrong? Power fails in the middle of the upgrade and the computer won't boot anymore? You just switch the root file system over to the snapshot and continue from there. That's literally just choosing a different entry in the boot manager.

fsck should take care of that to my knowledge.

fsck does not check data. At all. It only checks the file system's meta data for consistency. It can detect and hopefully fix errors in the file system itself, but it doesn't do anything for your data. Ext4 simply doesn't have the information needed for allowing such repairs. You try to play it safe by storing data on RAID 1? Good. If a disk reports an error, the RAID system can throw the disk out and alert you. What if your memory is faulty and the system writes wrong data on one of the RAID disks? Then you got a 50:50 chance of reading incorrect data. Nothing will detect this, nothing will alert you. And even if you scrub your RAID and detect the inconsistency, which of the two versions of your data is correct? There's no way to know. Btrfs on the other hand does keep checksums both of meta data and data and will detect errors and can use this information for choosing the correct version.

[coder]

niner I don't know, I don't use snapshots, actually, this is first time I did hear about that concept, and after reading for a bout 5-10 minutes about it I failed to see any benefit over traditional data backup

BTRFS snapshots are atomic, virtually instantaneous, and avoid duplication. Running hourly snapshots is an easy protection against accidentally deleting/clobbering a file. Then, when you want to make a backup, you can use the most recent snapshot to minimize inconsistencies.

And where package rollbacks won't help you is in relation to config files. /etc/ is one place I always enable snapshots.

Sub-volumes are another feature I quite like, though I concede it's largely motivated by my use of snapshots. It's a nice alternative to being forced to statically partition.

Now, one big downside of BTRFS is in COW and the penalty you take with large, frequently-modified files. Things like databases and VM images, for instance. So, I tend to use XFS to hold such data.

I know it's a lot to ask, but it's generally a good idea not to trash something you don't know/understand. It's fine to question the value/worth/usefulness of something, though. Questions are a good learning tool.

[leipero]

caligula Maybe because I do not care for such features? Why wasting time on reading something I (and 99% of the users) couldn't care less.

niner Ok, I see, still that's very unlikely scenario, and running something for the scenarios that are unlikely to happen (in years, it never actually happened to me on two separate machines, not counting others I've installed) it just doesn't make sense for desktop users (where my point was), for business it does, but I wasn't refering to that at all.

My understanding was that fsck checks data/blok integrity of file system, if file gets corrupted due to the bad RAM it is likely it will show in file system blocks on ext4, it's funny because that happened to me one week ago, when moving PC etc., one RAM module didn't have good contact, and I've got corrupted file system (and booting with half of RAM, two modules, sometimes it boots both sometimes just one well seated module...), did run fsck manually, everything got fixed and journalctl stopped reporting errors and system booted fine (after cleaning and re-seating RAM module ofc.).

coder Ok, I didn't trash it tho, I did question value for desktop (maybe in a bad manner...), because I still see none. Now for sure, people tend to clasify games/media and other nonsense as "important data", so they end up with terabytes of backups, personally I clasify that sort of things as irrelevant trash and end up with needs to backup few megabytes at best. For people who take lot's of family photos etc., it's not a good idea to save them in digital form only, and at one place only, so snapshots are questionably useful in those cases also.

[waxhead]

Michael: In your graphs - Is it possible to make some group boxes around tests that should be grouped e.g. btrfs related, xfs related, etc... etc.... I think that would make it a bit easier to follow.

[microcode]

Ok, then please tell me how I can have lots of snapshots with ext4 without a large slowdown. And how I can get easy rollback of updates without losing my data. And what about fast incremental backup both local and off-site? And how can I be sure with ext4 that my data is correct and not corrupted by a faulty disk or memory?

Just don't put yourself in a position to "need" these features. I have used EXT4 on a workstation every day I've worked for the last five years. From a professional perspective, I need accessible offsite backups anyway, so there is no point in relying on some btrfs-specific metadata format to accomplish that. I just use normal backup software and I don't have to worry about bugs in advanced filesystem features rendering my backups inaccessible. For almost every operation except appending, CoW snapshots are no more efficient than full revision backups.

[Michael]

Michael: In your graphs - Is it possible to make some group boxes around tests that should be grouped e.g. btrfs related, xfs related, etc... etc.... I think that would make it a bit easier to follow.

Currently there is no way. I would need to do some thinking for how to come up with the logic for doing so since all the graph generation is fully automated.

[waxhead]

Currently there is no way. I would need to do some thinking for how to come up with the logic for doing so since all the graph generation is fully automated.

Alrighty then. I would appreciate if you consider doing something like that or add a "weak background tint" per group. I think it would make the graphs easier to follow (especially when drunk)

[pal666]

It would be great to see some ZFS benchmarks as well, as it's the only other filesystem with a comparable feature set to btrfs.

Performance isn't everything.

actually it is not comparable. its design requires raid and this was single-drive test