Purpose:

A five point bonus will be awarded to the person who shows the best speedup on 8 and 16 processors in combination with the fastest time. Another five point bonus will be given to the person who is able to sort the most numbers and in the fastest time.

Coding in MPI:

Program the ODD-EVEN Transposition Sort to sort any number of integers ( at least 2 ), using a single "sorter" process per processor. It will probably be easiest to debug if you start by trying to sort 8 numbers on 8 processors.

To make it easier both for you to debug and me to grade, let's agree on a convention for input/output control. Order matters.

1. Where will the input come from
   • stdin - stdin or redirected file
     
   • random - randomly generated
     
2. What kind of output do we want
   • IO - Input numbers and sorted output
     
   • Sorted - Sorted output list only
     
   • TimingOnly - Timing only
     
3. How many numbers to sort (Should be at least 2 - with random only)

mprun -np 8 OE stdin IO
mprun -np 8 OE stdin Sorted
mprun -np 8 OE stdin TimingOnly
mprun -np 8 OE random IO 8
mprun -np 8 OE random Sorted 8
mprun -np 8 OE random TimingOnly 8

Output:

What to Submit:

1. The hardcopy should include a listing of your program,
2. A discussion of the algorithm you implemented,
3. A timing study of problem size versus number of processors, This should include charts and graphs of sorting various data set sizes (try at least 100, 1000, 10000, 100,000, 500,000, 1,000,000 and 10,000,000) on various numbers of processors (1 - 8 on parasite/paradise; 1-16 on paranoia) Since the metrics often used in parallel computing are speedup and efficiency, these should also be calculated and plotted. (Please note that traditionally, the number of processors is plotted on the x-axis and the time/speedup/efficiency are on the y-axis.) Discuss the results of your timing study.
   NOTE: You SHOULD be able to achieve speedup!
4. Contrast and compare programming in mpi, parallaxis, and linda. (You might consider doing this in table format.) Include items such as SIMD/MIMD, memory type, ease of programming, flexibility in the types of programming that can be handled, whether or not topology must be explicitly specified, and did the programmer have to be concerned about synchronization, termination and deadlock issues in your discussion.
5. Compare your experience of executing c-linda and mpi on a parallel machine. Which gave you the most speedup? Discuss why you think this was the case.

You are to electronically submit a directory called lab3dir using the command:

submit -v ncs-grd 531lab3 lab3dir

submit -v ncs-grd 735lab3 lab3dir

The grading sheet used is available here.

SETUP	{ Each  needs  as local A, and n. } 
	{ SETUP is an inherently sequential procedure. } 

for  to  do 

  IF  then 

    if (  ) then 
      { Odd-even exchange;  Get value from the adjacent  
        processor } 
      RECEIVE( ProcID + 1, Temp ) 
      SEND( ProcId + 1, MAX( Temp, A ) ) 
      { Keep smaller value } 
      A  MIN( Temp, A ) 
    endif 


    SEND( ProcId - 1, A ) 
    { Retrieve larger value } 
    RECEIVE( ProcID - 1, A ) 

  else 

    { Even-odd exchange; steps as above } 

    if (  ) then 
      SEND( ProcId - 1, A ) 
      RECEIVE( ProcID - 1, A ) 
    endif 

    if (  ) then 
      RECEIVE( ProcID + 1, Temp ) 
      SEND( ProcId + 1, MAX( Temp, A ) ) 
      A  MIN( Temp, A ) 
    endif 

  endif 

endfor 

OUTPUT		{ Inherently sequential }