I have some surprises with time measurements and would appreciate a reference to a statistic sheet or an article on the subject. I am doing a test in the fastest configuration possible : two PCs connected with a cross-cable. The interface is a 100BT Ethernet that can transfer theorically a maximum of 12,5 Mega bytes per second. The test consists of a little code that sends packets of 65.500 bytes (the length parameter of the WriteFileEx). The client marks the packet at its beginning (each packet a # mark), the server copies the mark at its end and the client when it receives the same packet compares the two marks to verify that everything has been tranfered. The client and the server use overlapped/IO. In this configuration, it is memory to memory without disk access on either side.
With 4 processes on the client side, that is 4 connexions, the test reaches a transfer rate of 17 Mega bytes per second, moving 65 Mega bytes (4 x 250 x 65500).
Did I missed something or is the cable faster (+36%) than its nominal tranfer rate under these conditions