next up previous
Next: Acknowledgements Up: Parallel Algorithms for Previous: Merging phase

Implementation Notes

 

Note that the performance graphs for the CM-5, Figures 3, 6, 10 - 17, are for SPLIT-C (version 1.2) programs linked with the CM-5 CMMD Message Passing Libraries (version 3.2), Figures 10, 18 - 19 are for the IBM SP-1 with the message passing libraries MPL, and Figures 7, 20 - 21 are for the IBM SP-2 with wide nodes and also MPL. Figures 8 and 10 are run on a Meiko CS-2 with SPLIT-C linked with the Elan Widgets message passing library. Note that our port of SPLIT-C to the CS-2 results in less than optimal performance because this SPLIT-C installation has not been fully optimized to make use of Elan, the low level communications library. We expect results using an optimized platform shortly. Figure 9 is implemented on an 8-processor Intel Paragon, using the PAM message passing libraries, the Paragon Active Messages platform from UC Berkeley.

The source code for the parallel algorithms presented in this paper, image histogramming and connected components, is available for distribution to interested parties.



next up previous
Next: Acknowledgements Up: Parallel Algorithms for Previous: Merging phase



David A. Bader
dbader@umiacs.umd.edu