Hi,
Thanks for your kindful comment.
My test (stress) environment was
- tool : tsung (with about 5 linux boxes to generate load)
- concurrent usrers: 10,000 GET operations (target url objects are
1,000 10KBs)
The reason why I use the completion callback was to make access log
compatible with apache format.
In libmicrohttpd manual, I could not find a proper API.
In normal operation, I used MHD_create_response_from_callback().
Although I can add my completion callback to it, some responses had to be
made from MHD_create_response_from_buffer().
In these cases, I could not add my completion callback to it.
That's why I used the MHD completion callback.
After a little optimization, my caching proxy can handle up to
10,000~20,000 GETs/sec (without keep-alive) in 2 xeon CPU (8 cores)
In the next phase, I'm going to measure elapsed times in the connection
handling of libmicrohttpd.
If you have any comments, I will be highly appreciate for that.
-------
weon
