The release is kinda strange in that it was the first in which the main changes were for performance. OK there is an improvement to resilience in the face of failed re-keying which some users were seeing in high load situations, but that high load was (in some cases) being caused by the daemon itself.
The process was mainly driven by one of our users, Nix, who was experiencing ekeyd using "too much" CPU on his system.
Of course on our servers during testing ekeyd it had used around a percent of CPU, certainly nothing that flagged as a problem in our own use (yes we eat our own dogfood ;-) Alas for this user on a 500MHz geode it was guzzling down 10% of his CPU which was clearly unacceptable.
This user however instead of guessing what the problem might be or simply leaving it up to us did something about it. He instrumented ekeyd, located the garbage collector tuning parameters as being incorrectly set and supplied a patch. Did he stop there? no! he then went on to profile the code further and clean up the hotspots. This resulted in ekeyd falling to less than 1% of the runtime of his system.
By reducing the CPU usage of ekeyd to this level it became more apparent where a previously reported bug was coming from, which enabled me to address it.
I know sometimes I complain about Open Source software but at times like this it makes me happy that we released the ekeyd software freely. This is how its supposed to be! Everyone working to make better software and benefiting together.
It has not just been on this occasion either, throughout the last year since our very first 1.0 release there has been helpful and useful feedback, patches from several users and even the odd thankyou mail. This project then has been a positive Open Source experience and I look forward to another constructive year maintaining this software.
No comments:
Post a Comment