Jim Rees wrote:
I wonder if the fsync is really necessary. This would go a hundred times
faster without it.
200 times on a standard RAID. Measured.
While I don't believe it's necessary, I did not dare to remove it
entirely, believing in the good reasons that must have motivated
somebody to fsync in the first place.
Instead, I came up with a patch that batches all the fsyncs together
into one on a suitable scale. We've been running like that for over a
year now with no ill effects. On the contrary - it motivated migrating
to the namei file server instead of the inode one which will still do
this inside the file system through inc()/dec().
I'll dig that one out and submit it.
--
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland
Phone: +41 22 767 8985 Fax: +41 22 767 7155
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel
