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

[pal666]

journaling is causing 2x writes on every metadata operation with no pay off.

pay off is no fsck on unclean shutdown

[pal666]

Oracle recently talked about changing ZFS' license so it could get in the Linux kernel.

someone from oracle, not oracle. certainly not oracle lawyer

[pal666]

If i'm not mistaken, ZFS will never be mainlined over licensing incompatibility...

nobody needs zfs in kernel anyway, its design is obsolete

[pal666]

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

now take another 5-10 minutes and read about atomicity

[pal666]

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.

you can disable cow per-file or per-subtree with chattr +C, no need to have separate partition

[pal666]

Maybe because I do not care for such features?

nobody cares for things he doesn't understand

My understanding was that fsck checks data/blok integrity of file system

your understanding was incorrect. on most filesystems fsck could only check metadata. btw, do you care about data/metadata distinction?

[pal666]

I need accessible offsite backups anyway, so there is no point in relying on some btrfs-specific metadata format to accomplish that.

it happens that there is a point and btrfs supports exactly that with send/receive

For almost every operation except appending, CoW snapshots are no more efficient than full revision backups.

snapshots are instantaneous and atomic. even for your backups it is better to make snapshot and backup snapshot

[coder]

Wow, you're on fire today.

you can disable cow per-file or per-subtree with chattr +C, no need to have separate partition

Thanks for that!

[leipero]

pal666 Well, both metadata and data are forms of data..., information is when it is new, as soon as it is known it is no longer information but data. Metadata describes relevant information about data, both are concepts, it's just a matter of semantics. From my understanding, modern file systems are smart enough that in case if data gets corrupted, metadata is updated about corruption after file system access/checks. So by checking metadata, fsck "checks" data too (sort off). In some (again very unlikely) scenarios corrption might happen in the way that metadata does not get refreshed, but my point was that improbable possibility those scenarios simply do not have benefit for desktop users to drag around more unnecessary complex file systems that in their nature should be as basic as possible to avoid security and other risks (to add, in other words, those "features" introduce complexity that rise a chance of data corruption at the end).

Just my non-informed logical view.

[pal666]

Just my non-informed logical view.

why don't you inform your misconceptions?