*** Niklas Böhm [2022-08-28 20:30]:
>That makes sense, then I guess this could be further decreased by setting
>REDO_NO_SYNC=1, although that was just something I noticed and am not
>bothered by.

Nothing is written during OOD-check, so NO_SYNC won't help in that case.
But during the build, when .redo/*.rec are created -- it will affect
filesystem operations.

>Thank you, that resolves my issue and helps by a lot.

Great, glad to hear about that hugely reduced time! There were huge
amount of syscalls, that are relatively slow especially because of
network filesystem delays.

>I agree that this is probably the explanation of the remaining difference,
>but that is an acceptable trade-off since the difference is not that large
>anymore.

I think it is acceptable tradeoff now too. Maybe there will be another
simpler format, but its worktime will be anyway negligible on NFS.

>By using separate files goredo plays much nicer with distributed
>systems; the sqlite database gets corrupted when more than one computer
>accesses it via a networked FS.

Somewhere apenwarr wrote that he also thinks his decision of single
database was wrong and database-per-directory should be preferrable.

-- 
Sergey Matveev (http://www.stargrave.org/)
OpenPGP: 12AD 3268 9C66 0D42 6967  FD75 CB82 0563 2107 AD8A