On 01.10.2010 15:56, Ramkumar Ramachandra wrote:
Hi Stefan,
Stefan Fuhrmann writes:
You obviously don't have APR threads enabled.
Thanks for finding this. Fixed in r1003430.
I enabled it, but there's still some issue:
subversion/svnadmin/main.c:1892: undefined reference to
`svn_fs_get_cache_config'
It builds here. Did you run autogen.sh before ./configure?
The server caches almost everything. So, the
first ("cold") run demonstrates the worst-case,
the second run ("hot") the best case. Real
world performance depends on server memory
and load.
Ah. Thanks for the clarification.
For the merge part: please review the "integrate-membuffer"
branch (only 3 patches). You may also review and
merge any of the patches listed in /branches/performance/STATUS.
integrate-cache-membuffer, you mean? Thanks! Your emails contains
references exactly to the patches I'm looking for :)
For the MD5 stuff, try r986459 (performance branch).
It should eliminate 1 of the 3 MD5 calculations.
Why doesn't STATUS document this and everything else consistently?
Because there is no simple mapping rev->feature / improvement.
In particular, there are a number of intermediate steps that were
replaced by new code later on. There is no point in reviewing
these earlier revisions but the newer ones can't be reviewed and
merged on their own. Hence the integration branch for the first
major change.
As soon as a larger number of patches got reviewed and merged,
I will prepare further changes for integration. So far, nobody had
free cycles to do the reviews.
As for the temp files, there are some improvements
listed in /branches/performance/STATUS. These would
also reduce your "system" CPU time.
I had the chance to check them out and apply them just now. They work
as expected. I'll submit some (uneducated) patch reviews to the list
and request a merge. I don't have access to the areas your patches
touch.
I really appreciate that. It would be great if someone had the time
to review the 3 commits to the membuffer cache integration branch.
The review should not require too much context knowledge. An
in-depth review will take a full day or so (like for an average sized
C++ class).
-- Stefan^2.
