Have you accounted for the possibility of clients falling off the map / buffers overflowing ?
Can you reduce the code to the simplest possible example that reliably demonstrates the problem and post it somewhere ?
On 2012-05-08, at 7:48 AM, Shahid wrote:
I have been using libevent in all my linux networking projects due to its features and AIO handling. I have a situation here and need your help. I have developed a PUSH (Comet) server using longpolling over http. I have used "libevent 2.0.14" for basic event handling and for http support. I have tested the server with 250 simultaneous connections and it performed well on a local\private network. When I tested it on WAN with just 10 browsers doing the long-polling with it, after 20 mins the raising of events from within libevent starts getting delayed. The application that generates PUSH Messages for this server is generating 220 messages per minute. That do not seem like a huge traffic for libevent, The delay is in raising the event from within libevent and it gets to 15 secs sometime. Please help me out on this. As my project delivery date is due now and I still stuck with this issue, I know that libevent can do far better than this. It could be some configuration issue in my case or something else, i don't know. I am using "Ubuntu 11.10" and "GCC 4.6".