On Jun 18, 2009, at 3:09 PM, Mark Phippard wrote:
> On Thu, Jun 18, 2009 at 12:08 PM, Johan
> Corveleyn<johan.corveleyn_at_uz.kuleuven.ac.be> wrote:
>
>> Anyway, I've been meaning to discuss this, because I think there is
>> a lot of room for improvement here, but now I just wanted to know
>> if anyone has seen "real-world" improvement by packing. Has anyone
>> ever compared timings of log and/or annotate pre- and post-packing?
>> That would be really interesting ...
>
> Packing was not supposed to improve performance. My understanding was
> that it is expected there would be a slight performance penalty as it
> would first try to open the rev file before opening the packed file.
Nope. We cache the extent to which a repository is packed, as well as
the various offsets within the pack file, so there are no additional I/
O accesses (once the offset cache is hot).
The claims of performance improvement stem mainly from increased use
of operating system disk caches, disk locality and so forth. In other
words, we're being smarter about giving the operating system the
ability to increase performance, but how the underlying OS chooses to
use that information could be quite arbitrary.
-Hyrum
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=1065&dsMessageId=2363317
To unsubscribe from this discussion, e-mail: [users-unsubscribe_at_subversion.tigris.org].
Received on 2009-06-18 22:17:16 CEST