Communication

Communication with traditional message-passing systems (like PVM or MPI) can be characterised as follows:

• data that should be passed have to be packed explicitly and type-dependently to a message into a send buffer
• for passing a message it is put in an envelope that carries delivery informations (sender and receiver identifications, message tag, message context)
• many different semantical variants for receiving: non-selective (any message from any sender), selective (certain message tag, certain sender), blocking, non-blocking etc.; a received message is always provided in form of a receive buffer reference
• the date have to be unpacked explicitly and type-dependently from a message

Typically, the above functionality is provided by a dozen functions.

Communication in the DP Model is characterised as follows:

• Data in MATLAB are always of type Array. MATLAB by itself provides powerful methods for building compound data objects (again of type Array). Therefore, message-passing can be simplified in MATLAB to pure array-passing.
• arrays can be send directly (optionally with an array tag as delivery information)
• all semantical receiving variants are provided by a single function, that returns the received array directly

The above functionality is provided by two functions.

Any kind of array can be sent to one or several DP instances (including the sender) with dpsend(array,dpids). If an array should be sent under a certain tag it can be provided as third parameter: dpsend(array,dpids,tag).

With dprecv arrays can be received selective (certain senders and/or certain tags) or non-selective. For detailed informations about dprecv see the online help or the reference guide.

Beside passing complete arrays, the DP-Toolbox supports also scattering and gathering of arrays. Scattering of arrays is possible with dpscatter(array,dpids,tag,dimension).

dpscatter divides the array along the specified dimension according to the number of destinations (dimension of dpids). That also works if the number of columns divided by the number of destinations gives a reminder. Scattered arrays may be received with dprecv.

The function A = dpgather(dpids,tag,dimension) receives an array from each source specified in dpids and builds them together into one array.

To test the functionality of dpsend/dprecv as well as dpscatter/dpgather functions two test applications are provided: dp_sendtest and dp_scattertest.