TCP/IP and Sockets for Control Objects

The MPI implements network functionality as client/server. The xmp\util\server.c program implements a basic server. You just create a Control object of type MPIControlTypeCLIENT and specify the server's host in the MPIControlAddress{}.client{} structure.

You can try "MPI networking" on a single machine by starting up the server program in a DOS window, and then running a sample application in another DOS window. Note that you can specify the host name and port of the server as command line arguments to all sample applications and utilities.

The way the MPI client/server works internally is that low-level mpiControlMemory and mpiControlInterrupt methods are intercepted just before they read/write XMP memory. The methods are packaged up as remote procedure calls and sent to the server for execution. The server sends the results back to the client.

There are 2 channels of communication - one channel to wait for interrupts, and another channel to do everything else. All MPI methods that communicate with the XMP do so by calling (eventually) the low level mpiControlMemory methods, so no application code needs to be changed other than the initial call to mpiControlCreate. This is all implemented on WinNT using WinSock.

Note that it would be possible to implement the client/server scenario above using an RS-232 line rather than TCP/IP WinSock. The MPI's client/server protocol only requires a reliable transport mechanism (WinSock, RS-232) between a client and server.

Return to Control Objects page