just was thinking as I crawled into bed, but dont want to forget it, so
typing it up now.

currently lookup searches for whiteout, even in bend.  Doesn't make
sense.

Double the amount of work lookup() has to do for instance in a opaque
dir that exists in just the top level branch.

similarly, in create, you search for whiteout, if first=last (single
level or opaque) waste of time.

unlink is the same, if dbstart=dbend (because of opaque or single level)
just do plain unlink, no need for partial_lookup() (I think, less
positive about this one).

probably other places, but need to go to sleep now, this could probably
help with benchmarks that operate in their own dir, especially where
lookup is slow (say network file systems).

too bad I just realized this and no time to code for usenix, wondering
if this would improve overhead I see for bonnie/postmark when unionfs is
using nfs as backing store (as nfs has significantly more overhead than
local ext3)

_______________________________________________
unionfs mailing list
[email protected]
http://www.fsl.cs.sunysb.edu/mailman/listinfo/unionfs

Reply via email to