CSCI 3366
Spring 2004
Homework Two
           
Continuing our discussion of divide and conquer algorithms, we will write
a C-MPI program which will perform a simple parallel sort. The master 
process will generate an array of a given size (command line) of random
numbers to be sorted. Depending on the number of processors available, 
the array will be divided among the processors so that each has an equal
share to sort. You may use an ordinary bubble sort (since we wish to make
the processors work hard in order to gain some benefit from parallelism).
The processors will sort their chunk of the array, and pass the sorted 
array back to the master process, who will display the sorted results. It 
is the master process's responsibility to merge the sorted chunks to provide
the original array in sorted order. (However, since bubble sort is O(n^2)
and merge is O(n), the bubble sort dominates the time taken for the sort.)
Using MPI_Wtime(), time how long it takes to sort various array sizes using
a varying number of processors, and see what you can conclude. I recommend
files of size increasing by an order of magnitude each time, such as 1000,
10000, 100000, 1000000, and perhaps some inbetween. You may also try 1, 2,
4, 8, 16 processors, etc and maybe note the machine speeds as well. 

Homework due Tuesday, February 3, 2004

int Bubble(int *a, int n)
{	int 
		notstop, 
		i;
	notstop = 1;
	while (notstop == 1)
	{	notstop = 0;
		for (i=0; i<n-1; i++)
		{	if (a[i] > a[i+1])
			{	int t;
				t = a[i]; a[i] = a[i+1]; a[i+1] = t;
				notstop = 1;
			}
		}
		n--;
	}
}