EDIT: No matter, figured it out. Seems it's the console output consuming all that time... reminds me why I don't like using the std lib. (Freely move to the trash can :))


Either I'm implementing them wrong, or they're extremely inefficent.

I did a simple test of transfering 2MB locally with both an IOCP and standard recv(). It took the recv() under 16ms, while the IOCP took 1,500ms+. I must admit I think I'm implementing IOCP wrong, I would think it's impossible to take anywhere near a second to receive data from a local endpoint.

NOTE: This is my first attempt using IOCP, one of the Window's concepts I never delved into till now.


(IOCP)
[code]
for(;;)
{
dwStartTime = GetTickCount();

if(WSARecv(hClient, &Buf, 1, &dwReceived, &dwFlags, &Overlapped, NULL) == SOCKET_ERROR)
{
if((dwError = GetLastError()) != 997)
{
std::cout << "Error = " << GetLastError() << std::endl;
}
}

else
{
if(dwReceived)
{
std::cout << "Immediate completion (dwReceived = " << dwReceived << ")\n";
dwTotal += dwReceived;
}
}

dwOther += GetTickCount() - dwStartTime;


dwStartTime = GetTickCount();

// wait
if((dwEnd = GetQueuedCompletionStatus(IoPort, &dwReceived, &dwKey, &pOverlapped, INFINITE)) != 0)
{
std::cout << "Async Completed (dwReceived = " << dwReceived << ")\n";
}

else
{
dwTotal += GetLastError();
}

dwMaxTime += GetTickCount() - dwStartTime;



if(dwTotal >= 2000000) break;
}
[/code]

(Standard recv(). (Synchronous/blocking)
[code]
for(;;)
{
dwTotal += recv(hClient, (char*) Data, 20000, NULL);

if(dwTotal >= 2 000 000) break;
}
[/code]

Excuse the sloppy code, just testing.

Note that, it seems WSARecv(...) is consuming most of the time (1,300ms+).