Algorithm 5. Pseudo-code of the Swap Phase for Parallel PAM

1. repeat

2.  for p := 1 → P do in parallel

3.   for each data point Xi ϵ p do      /* All PEs do in parallel*/

4.    for j := 1 → K do

5.     Find Euclidean distance b/w Xi and Mj

6.     Tag Xi with j for which this distance is minimum

7.    endfor

8.   endfor

9.   endfor

10.  Concatenate clustering results from all PEs      /* All PEs do this*/

11.  for p := 1 → P do in parallel

12.   for j := 1 → K do        /* All PEs do in parallel*/

13.    for each data point Xa ϵ p & cluster j do

14.     for each data point Xi ϵ cluster j do

15.      Find sum of Euclidean distances b/w Xa and Xi

16.     endfor

17.     Find minimum sum for each cluster

18.    endfor

19.   endfor

20.  endfor

21.  for j := 1 → K do        /* Only master PE will do this */

22.   for p := 1 → P do

23.    Update Xa as jth Medoid for which the sum is minimum from all PEs

24.   endfor

25.  endfor

26. until no update in any of ‘K’ Medoids