[Borgbackup] Performance of using Borg on SMR drives

[Marian Beermann]

I saw a long thread in the Linux Kernel Bugzilla, some versions of the
kernel apparently had a lot of problems with device-managed SMR (IO
errors, timeouts, unresponsive devices when writing more than a couple
GB), but it's supposed to be all fixed now. (If I understand correctly
some people still see these issues with some workloads, but "in general"
it seems to work -- more or less?)

https://bugzilla.kernel.org/show_bug.cgi?id=93581

On 03.03.2017 15:39, John McCabe-Dansted wrote:
> I haven't got borgbackup to work well on SMR drives. I get hardware
> errors and a stale lock file 100GB or so, but I have only tried 1.0.x.
> 
> Maybe I should try again with 1.1.x
> 
> On 3 March 2017 at 06:07, Marian Beermann <public at enkore.de
> <mailto:public at enkore.de>> wrote:
> 
>     On 02.03.2017 11:59, Marian Beermann wrote:
>     > Did someone try this yet?
>     >
>     > From reviews like
>     > http://www.tomsitpro.com/articles/seagate-8tb-archive-hdd-review,2-822-3.html
>     <http://www.tomsitpro.com/articles/seagate-8tb-archive-hdd-review,2-822-3.html>
>     > I take it that the banded nature of a device-managed SMR drive means
>     > that low-concurrency sequential writes and no overwrites are best for
>     > these drives -- just like Borg does.
>     >
>     > However, I think the 5 MiB segment size (+fsync) used in 1.0.x might be
>     > very bad for performance here (since the bands are larger than 5 MiB).
>     > 500 MiB (1.1.x) works maybe better there.
>     >
>     > Cheers, Marian
>     >
> 
>     ^- I ordered one now to try it, the Samsung Archive one which is
>     device-managed SMR.
> 
>     We'll see.
>     _______________________________________________
>     Borgbackup mailing list
>     Borgbackup at python.org <mailto:Borgbackup at python.org>
>     https://mail.python.org/mailman/listinfo/borgbackup
>     <https://mail.python.org/mailman/listinfo/borgbackup>
> 
> 
> 
> 
> -- 
> John C. McCabe-Dansted