This directory contains programs which use collective communication functions.
In this scenario, we would like to send some information from the master process to all the other processes in the communicator. This situation could arise if the master process had read some parameter settings from file, or it had received some information from one or more other processes and needs to coordinate the rest.
This program contrasts the time taken to achieve this using a series of point-to-point communications against that used by one collective broadcast operation.
Run the program several times using different numbers of processors and plot how the communication times scale over an increasing number of processors in the communicator.
Another common scenario is one in which each process has computed a value which we would like to combine in some way: add, multiply, divide etc.
In this program we revisit the trapezoidal approach to numerical integration. Instead of sending the subtotals back to the master processes as a sequence of point-to-point communications, we can call a collective reduce operation to achieve the same result, but more efficiently.
This last program demonstrates two more collective operations: scatter and gather.
The call to scatter chunks up a send buffer on one of the processes (called the root) and distributes those chunks to all the other processes.
The call to gather takes chunks from all the other processes and collates them into a single buffer on the root process, i.e. it performs a mirror image of the scatter operation.
Our master send buffer contains a line from Macbeth, which is 66 characters long. Pay close attention to the contents of the reconstituted buffer as you vary processor count.