[kwlug-disc] BTRFS Was: Transferring large backup files over the network

[CrankyOldBugger]

Not to heat up an argument or anything, but as coincidence would have it,
this article was in my news feed this morning and I thought it might be
relevant.  I don't know if the writer used credible testing methods or
anything, nor do I endorse his opinion in anyway, etc...

https://www.ctrl.blog/entry/btrfs-vs-ext4-performance.html



On Fri, 24 May 2019 at 00:21, Chris Irwin <chris at chrisirwin.ca> wrote:

> On Thu, May 23, 2019 at 10:06:32PM -0400, Remi Gauvin wrote:
> > many places, including the official
> >BTRFS wiki, recommend disabling Copy-On-Write for database or VM Image
> >files.
>
> I use BTRFS for my desktop and laptop (both home and work) and have no
> issues or complaints. Automatic snapshots have saved me several times.
>
> However, BTRFS is a tool. You can use it for the features it was
> intended for (COW and checksumms), or you can use it wrong. Anything
> suggesting disabling COW on BTRFS is simply plain wrong, regardless of
> whether it is some blog or official documentation.
>
> Disabling COW for VMs on BTRFS means:
>
>     * You're not getting data checksums (they require COW)
>     * You're not using snapshots (they silently enable COW)
>
> If you're not using either of those: Why are you bothering with BTRFS?
>
> BTRFS with COW disabled is worse than ext4: You're not getting any
> actual feature benefits, and you've lost the more plentiful and robust
> ext4 recovery tools (should you ever need them).
>
> Additionally some features (like snapshots) will silently enable COW,
> even on files that COW was "disabled", and defragmenting (including
> autodefrag) will cause data duplication between snapshots (which were
> deduped by virtue of COW).
>
> For Virtual Machines and Databases, I'd really just recommend having an
> ext4 LV for those. I keep VMs and databases on ext4 (at home) or xfs (at
> work, due to RHEL).
>
> (I prefer ext4 use for home, because you can't shrink xfs)
>
> >However, the BTRFS raid implementation is completely dependent
> >on CoW to maintain synchronization between mirrors. Any event that
> >interrupts writes to a mirrored file with CoW disabled will result in
> >discrepancies between the mirrored copies, and they are never
> >automatically fixed.. not even if you run a scrub.
>
> Which makes sense. Disabling COW disables data checksums, and scrub
> searches for checksum errors.
>
> --
> Chris Irwin
>
> email:   chris at chrisirwin.ca
>  xmpp:   chris at chrisirwin.ca
>   web: https://chrisirwin.ca
>
> _______________________________________________
> kwlug-disc mailing list
> kwlug-disc at kwlug.org
> http://kwlug.org/mailman/listinfo/kwlug-disc_kwlug.org
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://kwlug.org/pipermail/kwlug-disc_kwlug.org/attachments/20190524/72ba4bf5/attachment.htm>