serious problem in send()

[Damon123]

I am implementing multi-server-client program.

I use the select function to detect sockfd which is available to read and write so as to solve the blocking problem when send/recv.

Now, I meeting a interesting problem.

Although the select() indicate a sockfd ( Server B) is available to be written (there is send buffer left in the kernel), when I call send() to send a large packets (which the size is larger than the remaining kernel buffer size) to Server B, it may still be blocked if Server B is busy and not calling recv().

Then i use setsockopt to set send timeout. But I found that since the large packets is divided into various packets and already sent out a few packets. After timeout it return error, but it cannot discard those packets already sent out. Server B then will receive incomplete msg.

How can I ensure check the current remaining kernel buffer size before I start sending out the first packets ?

[MikeAThon]

Don't send a single large packet. Send multiple smaller ones untile send() returns WSAEWOULDBLOCK (use non-blocking sockets)

Mike