On Sep 6, 2005, at 1:43 PM, Ryan Schmidt wrote:
> On Sep 6, 2005, at 21:25, Paul Koning wrote:
>> I'm doing more testing on my newly converted repository (25,000
>> revisions, not 19,000 as I said earlier).
>> Right now it's running a checkout via svnserve. That was running
>> nicely for a while. But right now it has slowed to a crawl -- about
>> 30 seconds per file, doing small files. svnserve seems to be idle,
>> there is hardly any disk traffic -- nothing at all to explain why
>> is happening. Server is on Mac OS 10.4.2, client on NetBSD.
>> Any ideas? Should I try a BerkeleyDB based repository?
> I can't recommend that.
> Mac OS X should be very capable of serving or using Subversion
> In addition to the issue Stephen mentioned, there is a performance
> issue on some operating systems when APR is compiled to use /dev/
> random instead of /dev/urandom. Unfortunately I believe this is the
> default, as it's how it ended up on my system when I installed
> Subversion via DarwinPorts:
> $ cd /opt/local/lib
> $ grep /dev/random libapr-0.0.dylib
> Binary file libapr-0.0.dylib matches
> $ grep /dev/urandom libapr-0.0.dylib
> I do not know if Mac OS X is affected by the problem, but if you
> find that your APR also uses /dev/random, you can try changing it.
> There's a switch you pass to the ./configure program to tell it
> which source of entropy to use. If your copy of APR is built
> specially for Subversion, then you can pass this option to
> Subversion's ./configure and it'll pass it on to APR. Else pass the
> switch to APR's ./configure directly when you build it.
Mac OS X is not affected by the /dev/random issue.
Received on Tue Sep 6 22:54:11 2005