With one of my hlfs systems running an active subversion repository, I noticed immediately a hang issue with all client to the server. After some lengthy searching I believe that I isolated it to a single issue: entropy.
The problem here is that subversion uses apr (Apache Portable Runtime) whether you use or want apache or not.. The apr seems to abuse the /dev/random by default and absorb a large chunk of entropy during basic io to and from the repository. The machine hosting the repository is dedicated to this for a small group of users (~10). Therefore the hardware requirements are not that huge. Unfortunately this machine cannot keep up with subversion and apr, and fill in the entropy while multiple users are trying to use this. (and subversion doesn't care, it just hangs and waits, so all users get confused thinking that they lost connection). Now, I am going to test this with and without the entropy_pool size patch that increases the entropy. The current system is running with that patch enabled, but due to it's slow specs, it never makes it very far. On average it sits around 200 when it should be around 1000 and at max 16384. Is it possible for the entropy pool size patch able to cause this slowdown? I'll also see if I made any non-standard changes (as this system is somewhat experimental) that may be causing this problem. (perhaps the read/write triggers are higher than they should be..) -- Kevin Day -- http://linuxfromscratch.org/mailman/listinfo/hlfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
