i have a question, whats more efficient:

- queue multiple WSARecv overlapped operations
- queue WSARecv only upon receiving FD_READ event


i guess that second part.
i dont know if its worth effort.

first approach will use constant ammount of memory, more connections = more memory


second approach will require me to spin though several bilions of clients calling function on each of them.


perhaps window queue method would be better?
at least i wont have to call WSAEnumNetworkEvents, just loop through socket list to get pointer to my data...

i dont know.

Use asynchronously posted WSARecv along with I/O completion ports and a thread pool (look at BindIoCompletionCallback). An I/O completion post will be equivalent to receiving an FD_READ event. There are plenty of examples on www.codeproject.com, though beware, some are subpar. You can also take a look at the win32 implementation of boost::asio, if you have the patience reading thru heavily templated code