Hi Danti,
away from my computer now, so I cannot check, but I suspect the --inplace to be
the culprit. If rsync has to overwrite the file while transferring, it may not
be able to check blocks in the original?
Regards
Hardy
--
I'm so stupid today, I could rule America.
On 29.07.25 12:55, Gionatan Danti via rsync wrote:> Hi all,
> rsync seems slow to copy large files. The issue is not related to the
transfer itself, which is quite fast, but to the discovery of different blocks.
>
> For example, transferring a big (multi-GB) file from src (remote) to dst
(local) with "--inplace" (to avoid a whole-file copy on dst) shows the
following:
>
> - if dst does not have a local file, or if -W was given, the copy
immediately starts (but it clearly needs to transfer all blocks);
>
> - if dst has a previous file version (and -W was not given), first dst
reads the entire file, then src does the same, next different blocks are copied
and finally dst re-read the entire file.
>
> As dst and src file scans do not happen concurrently, and including the
last whole-file read by dst, copy time is vastly increased at about 3x compared
to an optimized process (with concurrent scans and without the final read).
>
> Am I missing something? Can rsync be faster for large-file copies?
>
> Regards.
>