With thread-per-core model, each thread accepts connections by it own.
However, because the number of requests on each connection and the processing load of each request are different, the load situation of different cores will be different.
We expect the different core loads to be close so that the best overall throughput can be achieved.
With #87 we can analyze live connection counts of all threads, so we can implement an special accepter to make it possible to dispatch connections according to statistical data. It can alleviate the problem of #86.
TODO:
What statistical data to use? It can be live connection count or others(maybe provided by runtime).
With thread-per-core model, each thread accepts connections by it own.
However, because the number of requests on each connection and the processing load of each request are different, the load situation of different cores will be different.
We expect the different core loads to be close so that the best overall throughput can be achieved.
With #87 we can analyze live connection counts of all threads, so we can implement an special accepter to make it possible to dispatch connections according to statistical data. It can alleviate the problem of #86.
TODO: What statistical data to use? It can be live connection count or others(maybe provided by runtime).