The curves indicate that our proposed algorithm is perfectly parallel it gives approximately 80% efficiency. Figure 6 shows the efficiency curves for different problem sizes. The curves indicate that the speedup improves as number of processors increases. Figure 5 present the speedup curves on each processors. The exper- mental results show that the proposed parallel algorithm perform better for large problems. All the execution times are normalized relative to the sequential GS. Figure 4 present the real execution time (in seconds) of the parallel GS algorithm for different problem sizes run with 4, 8, and 16 processors. In the experiments we used the matrix orders of 32, 64, In the implementation process we used cyclic distribution functions. The workstations use Mandrake Linux 7.2 operating system running a kernel version 2.2. These hosts are connected together using 16-port Myrinet switch providing full-duplex 2 + 2 Gbps data rate links. this section, we evaluate experimentally the performance of the proposed algorithm on a cluster of sixteen Linux workstations each of which has a single Pentium IV with 128 MB of memory and 20 GB disk space.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |