* fixed the non keep-alive performance issue as well as i lower memory usage by using a single read and write circular buffer for every connection. @TODO: i noticed a server hang while getting large data (my image) with non keep-alive connections. Additionally an incomplete keep-alive request might stop the server now as the lock on the read buffer will not be released. (origin/master, origin/HEAD)
2012-02-19 15:41:48 +0100 Georg Hopp
* another try with the shmen trick...this time use MAP_ANONYMOUS ... as GNU extension.
2012-02-19 14:33:42 +0100 Georg Hopp
* Merge remote branch 'origin/master'
2012-02-19 12:13:52 +0100 Georg Hopp
* added missing header file to repo
2012-02-19 11:35:15 +0100 Georg Hopp
* another try with a shared memory based ringbuffer...this performs well for keep-alive sessions but is much slower without. actually i am not sure why but most likely the shared memory setup is quite expensive. @TODO: make a profiling. (HEAD, ringbuffer)
* another try with a shared memory based ringbuffer...this performs well for keep-alive sessions but is much slower without. actually i am not sure why but most likely the shared memory setup is quite expensive. @TODO: make a profiling.
2012-02-18 21:08:32 +0100 Georg Hopp
* fix inf loop. @TODO: This whole handling has to be cleaned.
2012-02-18 20:50:01 +0100 Georg Hopp
* this change hopefully makes the shm trick work on amd64
2012-02-18 20:12:27 +0100 Georg Hopp
* lots of changes but primarily change the request parser to use a ringbuffer. The ringbuffer is implemented using the shared memory trick.
