On 25.08.2011 10:09, Daniel Shahaf wrote:
> Paul Querna --- who wrote the initial revprop packing patch (the f5 one,
> which has been reverted) --- mentions that his use case for revprop
> packing has vanished due to hardware upgrades (acquiring SSD l2arc
> disks).
From my POV, there are still 2 issues that make
revprop packing worthwhile.
(1) Having a large number of (small) files can
become a nuisance if you want to copy or archive
repositories. Standard zip (and possibly other
archive formats) doesn't support > 64k files and
Windows people usually don't tar.
(2) Revprop lookup *is* a performance critical
operation because its result cannot be reused
across multiple SVN requests and many different
timestamps have to be fetched for c/o, export
and log (i.e. revprop access is a frequent operation).
Having fewer revprop files reduces file I/O.
-- Stefan^2.
Received on 2011-08-27 15:13:20 CEST