Hello,

I'm currently try to delete a directory with more than 50k
entries on a heavily loaded server. Wondering why this takes
so long I straced "rm" and found out that only 2..3 entries
are unlinked per second.

This really slow rate is, I suppose, due to the fact that
metadata updates are made disk-synchronous. This is, for my
application, not necessary. I can live with large metadata
changes batched into one large transaction.

Is it possible to switch off synchronous metadata updates
(while the filesystem is mounted) so that those changes are
reflected only some times later or when I switch off
asynchronous metadata updates?

Xu�n.


Reply via email to