<div class="gmail_quote">On Wed, Feb 16, 2011 at 12:54, Matija Kecman <span dir="ltr"><<a href="mailto:matijakecman@gmail.com">matijakecman@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div id=":d4">After cleaning up the log files and plotting log ( ||Ae||/||Ax|| )<br>
with iteration number I generated the attached figure. I am wondering<br>
why the number of iterations for convergence depends on the number of<br>
processors used? According to the FAQ:<br>
<br>
'The convergence of many of the preconditioners in PETSc including the<br>
the default parallel preconditioner block Jacobi depends on the number<br>
of processes. The more processes the (slightly) slower convergence it<br>
has. This is the nature of iterative solvers, the more parallelism<br>
means the more "older" information is used in the solution process<br>
hence slower convergence.'<br>
<br>
but I seem to be observing the opposite effect.</div></blockquote><div><br></div><div>You are using the same number of subdomains, but they are shaped differently. It seems likely that you have Parmetis installed in which case PCASM uses it to partition multiple subdomains on each process. In this case, those domains are not as good as the rectangular partition that you get by using more processes. Compare:</div>
<div><br></div><div><div><font class="Apple-style-span" face="'courier new', monospace">$ mpiexec -n 1 ./ex8 -m 200 -n 200 -sub_pc_type lu -ksp_converged_reason -pc_type asm -pc_asm_blocks 4 -mat_partitioning_type parmetis</font></div>
<div><font class="Apple-style-span" face="'courier new', monospace">Linear solve converged due to CONVERGED_RTOL iterations 27</font></div></div><div><div><font class="Apple-style-span" face="'courier new', monospace">$ mpiexec -n 1 ./ex8 -m 200 -n 200 -sub_pc_type lu -ksp_converged_reason -pc_type asm -pc_asm_blocks 4 -mat_partitioning_type square</font></div>
<div><font class="Apple-style-span" face="'courier new', monospace">Linear solve converged due to CONVERGED_RTOL iterations 22</font></div></div><div><div><font class="Apple-style-span" face="'courier new', monospace">$ mpiexec -n 4 ./ex8 -m 200 -n 200 -sub_pc_type lu -ksp_converged_reason -pc_type asm -pc_asm_blocks 4</font></div>
<div><font class="Apple-style-span" face="'courier new', monospace">Linear solve converged due to CONVERGED_RTOL iterations 22</font></div></div><div><br></div></div>