On Jul 15, 2006, at 21:28, Straw, Scott J wrote:
> Ryan Schmidt wrote:
>
>> gmu 2k6 wrote:
>>
>>> is there any performance impact whether I try to import an existing
>>> tree in one step by import or add/commit or if I do it in some
>>> well-defined steps? fsfs will be the backend.
>>> btw, while trying to import a >5GiB tree with many C/Java sources
>>> and
>>> a lot of binary resources svnserve (started via xinetd) just stopped
>>> and refused any connections (according to the error message on the
>>> client side). right now I'm trying to do it by adding and
>>> committing.
>>> is there anyway to get logs from svnserve?
>>
>> I don't think there are any logs from svnserve.
>>
>> Commits of > 2 GiB are supported if you use APR 1.2.x or greater.
>> They are not possible with APR 0.9.x for example. Check what version
>> of APR you're using.
>>
>> I have read recently on this list that there can be a performance
>> problem if you have a very large commit and then try to do something
>> with that revision, such as look at a log for it (IIRC). If these are
>> separate projects you're importing, you'd do better to import them in
>> separate commits.
>
> I hit that performance problem. Once we turned per-directory access
> control off (SVNPathAuthz off; see the Subversion doc, under Disabling
> Path-based checks), everything was hunky-dorey again.
Ok, but that's mod_dav_svn, and gmu2k6 mentioned he's using svnserve.
Do we know if a similar performance issue exists with svnserve? If
so, how would one turn off path-based authorization in svnserve?
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org
Received on Sat Jul 15 21:57:09 2006