<div class="gmail_quote">On Sun, Jul 22, 2012 at 9:22 AM, Bao Kai <span dir="ltr"><<a href="mailto:paeanball@gmail.com" target="_blank">paeanball@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Hi, Jed,<br>
<br>
The following is the output. Two equations are solved during each<br>
time steps. One is using bicgstab is already very fast, the other one<br>
is using gmres+ml.<br></blockquote><div><br></div><div>These equations are different? Why do you think they should take a similar amount of time?</div><div><br></div><div>Most of your execution time is spent in the 31 PCSetUp calls (not all of which are ML). If only a few iterations are needed, ML setup is not worth it. Also, if the system is not changing much, you could lag the preconditioner so that you pay for setup less frequently. This can be done at the time integration level (e.g. using a Rosenbrock(-W) method) or at the nonlinear solve by a variety of schemes including modified Newton, MFFD Jacobian, Quasi-Newton, or NGMRES.</div>
<div><br></div><div>Note that you can use PetscLogStagePush/Pop to profile each solve separately.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
The log_summary output can found in the end of the output.<br>
<br>
Best Regards,<br>
Kai<br>
<br>
> Message: 6<br>
> Date: Sun, 22 Jul 2012 07:16:07 -0500<br>
> From: Jed Brown <<a href="mailto:jedbrown@mcs.anl.gov">jedbrown@mcs.anl.gov</a>><br>
> To: PETSc users list <<a href="mailto:petsc-users@mcs.anl.gov">petsc-users@mcs.anl.gov</a>><br>
> Subject: Re: [petsc-users] How to investigate the reason for slow<br>
> convergence rate?<br>
> Message-ID:<br>
> <<a href="mailto:CAM9tzSkxKz5FwFJJPZz0UO5eN-S45QsdQxmo7xnH345R1EdTEQ@mail.gmail.com">CAM9tzSkxKz5FwFJJPZz0UO5eN-S45QsdQxmo7xnH345R1EdTEQ@mail.gmail.com</a>><br>
> Content-Type: text/plain; charset="utf-8"<br>
><br>
> Please send -log_summary output with performance questions.<br>
><br>
> On Sun, Jul 22, 2012 at 4:01 AM, Bao Kai <<a href="mailto:paeanball@gmail.com">paeanball@gmail.com</a>> wrote:<br>
><br>
>> Hi, Matt,<br>
>><br>
>> I tried ML6.2 with petsc3.3 with default parameters by only specifying<br>
>> the preconditioner to be PCML. The KSP solver is gmres.<br>
>><br>
>> The convergence rate is much faster, while it still took much longer<br>
>> time in total.<br>
>><br>
>> For example, for the problem with 500^3 mesh ( 125 million unknowns )<br>
>> with 512nodes(4 processors per node) on bluegene/P , it tooks about<br>
>> ten iterations to get convergent while the total time used is about<br>
>> 400 seconds.<br>
>><br>
>> 506 the KSP type is gmres<br>
>> 507 the PC type is ml<br>
>> 508 KSP rtol = 0.100000000000000008E-04 abstol =<br>
>> 0.100000000000000001E-49 dtol = 10000.0000000000000 maxit = 10000<br>
>> 509 SNES rtol = 0.100000000000000002E-07 abstol =<br>
>> 0.100000000000000001E-49 stol = 0.100000000000000002E-07 maxit =<br>
>> 50 maxf= 10000<br>
>> 510 0 SNES Function norm 5.859593121800e+02<br>
>> 511 0 KSP Residual norm 8.340827070202e+06<br>
>> 512 1 KSP Residual norm 7.980806572332e+05<br>
>> 513 2 KSP Residual norm 1.870896731234e+05<br>
>> 514 3 KSP Residual norm 6.790580947452e+04<br>
>> 515 4 KSP Residual norm 2.665552335248e+04<br>
>> 516 5 KSP Residual norm 1.130212349885e+04<br>
>> 517 6 KSP Residual norm 4.053599972292e+03<br>
>> 518 7 KSP Residual norm 1.786770710693e+03<br>
>> 519 8 KSP Residual norm 7.313571654931e+02<br>
>> 520 9 KSP Residual norm 3.205683714450e+02<br>
>> 521 10 KSP Residual norm 1.263243312734e+02<br>
>> 522 11 KSP Residual norm 3.945082815178e+01<br>
>> 523 1 SNES Function norm 9.378772067642e-02<br>
>> 524 0 KSP Residual norm 5.413489711800e+01<br>
>> 525 1 KSP Residual norm 1.442598710609e+01<br>
>> 526 2 KSP Residual norm 4.073537172140e+00<br>
>> 527 3 KSP Residual norm 1.157455598705e+00<br>
>> 528 4 KSP Residual norm 3.509855901968e-01<br>
>> 529 5 KSP Residual norm 1.160625342728e-01<br>
>> 530 6 KSP Residual norm 3.209351890216e-02<br>
>> 531 7 KSP Residual norm 7.780869881329e-03<br>
>> 532 8 KSP Residual norm 1.820828886636e-03<br>
>> 533 9 KSP Residual norm 4.172544590190e-04<br>
>> 534 2 SNES Function norm 6.747963806680e-07<br>
>> 535 Number of KSP iteration is 9<br>
>> 536 SNES solve takes time 406.724867261176314<br>
>><br>
>> But with bcgs and bjacobi, it tooks about 550 KSP iterations ( 2 snes<br>
>> iterations ) and 69 seconds to get the result.<br>
>><br>
>> For much smaller problems, benefiting from the fast convergence, it<br>
>> did takes less time to get the result. It seems that the ml can not be<br>
>> scaled, or I used it in a wrong way.<br>
>><br>
>> Best Regards,<br>
>> Kai<br>
>><br>
>><br>
>> > Message: 2<br>
>> > Date: Sat, 21 Jul 2012 08:47:10 -0500<br>
>> > From: Matthew Knepley <<a href="mailto:knepley@gmail.com">knepley@gmail.com</a>><br>
>> > To: PETSc users list <<a href="mailto:petsc-users@mcs.anl.gov">petsc-users@mcs.anl.gov</a>><br>
>> > Subject: Re: [petsc-users] How to investigate the reason for slow<br>
>> > convergence rate?<br>
>> > Message-ID:<br>
>> > <<br>
>> <a href="mailto:CAMYG4GkfV6kmTFEKXUadOv%2B2CrKHk9hRY7UR-cFWf%2BvcRxCv5g@mail.gmail.com">CAMYG4GkfV6kmTFEKXUadOv+2CrKHk9hRY7UR-cFWf+vcRxCv5g@mail.gmail.com</a>><br>
>> > Content-Type: text/plain; charset="iso-8859-1"<br>
>> ><br>
>> > On Sat, Jul 21, 2012 at 4:30 AM, Bao Kai <<a href="mailto:paeanball@gmail.com">paeanball@gmail.com</a>> wrote:<br>
>> ><br>
>> >><br>
>> >>> HI, all,<br>
>> >><br>
>> >> I am still suffering from the slow convergence rate of the KSP<br>
>> >> solution.<br>
>> >><br>
>> >> I changed the code to use Petsc3.3 and then try the gamg<br>
>> >> precoditioner,<br>
>> >> the convergence rate is better, while it took more total time because<br>
>> >> it<br>
>> >> took much more time for each iteration and some extra time for<br>
>> >> pre-processing.<br>
>> >><br>
>> ><br>
>> > Try ML to see if it has better defaults for your problem. If not, you<br>
>> will<br>
>> > have to start experimenting with the solver<br>
>> > parameters.<br>
>> ><br>
>> ><br>
>> >> I am wondering if there are some ways that can help me to investigate<br>
>> the<br>
>> >> slow convergence rate for KSP solution so that I can do some<br>
>> improvement.<br>
>> >> Is DMMG will be a good solution?<br>
>> >><br>
>> ><br>
>> > No.<br>
>> ><br>
>> > Matt<br>
>> ><br>
>> ><br>
>> >> Thank you very much.<br>
>> >><br>
>> >> Best Regards,<br>
>> >> Kai<br>
>> >><br>
>> >>><br>
>> >>> Message: 2<br>
>> >>> Date: Wed, 11 Jul 2012 15:17:15 -0500<br>
>> >>> From: Matthew Knepley <<a href="mailto:knepley@gmail.com">knepley@gmail.com</a>><br>
>> >>> To: PETSc users list <<a href="mailto:petsc-users@mcs.anl.gov">petsc-users@mcs.anl.gov</a>><br>
>> >>> Subject: Re: [petsc-users] Does this mean the matrix is<br>
>> >>> ill-conditioned?<br>
>> >>> Message-ID:<br>
>> >>> <CAMYG4Gk7T=<br>
>> >>> <a href="mailto:q%2Bw1PKO7G_TW07iDzux90Sncbv9K7d0FD-MDrLRg@mail.gmail.com">q+w1PKO7G_TW07iDzux90Sncbv9K7d0FD-MDrLRg@mail.gmail.com</a>><br>
>> >>> Content-Type: text/plain; charset="iso-8859-1"<br>
>> >>><br>
>> >>> On Wed, Jul 11, 2012 at 12:40 PM, Bao Kai <<a href="mailto:paeanball@gmail.com">paeanball@gmail.com</a>><br>
>> >>> wrote:<br>
>> >>><br>
>> >>> > Hi, all,<br>
>> >>> ><br>
>> >>> > The following is the output from the solution of a Poisson equation<br>
>> >>> > from Darcy's law.<br>
>> >>> ><br>
>> >>> > To compute the condition number of matrix, I did not use PC and use<br>
>> >>> > GMRES KSP to do the test.<br>
>> >>> ><br>
>> >>> > It seems like that the condition number keep increasing during the<br>
>> >>> > iterative solution. Does this mean the matrix is ill-conditioned?<br>
>> >>> ><br>
>> >>><br>
>> >>> Generally yes. Krylov methods take a long time to resolve the<br>
>> >>> smallest<br>
>> >>> eigenvalues, so this approximation is not great.<br>
>> >>><br>
>> >>><br>
>> >>> > For this test, it did not achieve convergence with 10000<br>
>> >>> > iterations.<br>
>> >>> ><br>
>> >>> > When I use BJOCABI PC and BICGSTAB KSP, it generally takes about<br>
>> >>> > 600<br>
>> >>> > times iteration to get the iteration convergent.<br>
>> >>> ><br>
>> >>> > Any suggestion for improving the convergence rate will be much<br>
>> >>> > appreciated. The solution of this equation has been the bottleneck<br>
>> of<br>
>> >>> > my code, it takes more than 90% of the total time.<br>
>> >>> ><br>
>> >>><br>
>> >>> Try ML or GAMG.<br>
>> >>><br>
>> >>> Matt<br>
>> >>><br>
>> >>><br>
>> >>> > Thank you very much.<br>
>> >>> ><br>
>> >>> > Best Regards,<br>
>> >>> > Kai<br>
>> >>> ><br>
>> >>><br>
>> >>><br>
>> >>><br>
>> ><br>
>> ><br>
>> > --<br>
>> > What most experimenters take for granted before they begin their<br>
>> > experiments is infinitely more interesting than any results to which<br>
>> their<br>
>> > experiments lead.<br>
>> > -- Norbert Wiener<br>
>> > -------------- next part --------------<br>
>> > An HTML attachment was scrubbed...<br>
>> > URL:<br>
>> > <<br>
>> <a href="http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120721/eaf7b2ee/attachment-0001.html" target="_blank">http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120721/eaf7b2ee/attachment-0001.html</a><br>
>> ><br>
>> ><br>
>> > ------------------------------<br>
>> ><br>
>><br>
> -------------- next part --------------<br>
> An HTML attachment was scrubbed...<br>
> URL:<br>
> <<a href="http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120722/7f782dd6/attachment.html" target="_blank">http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120722/7f782dd6/attachment.html</a>><br>
><br>
> ------------------------------<br>
><br>
> _______________________________________________<br>
> petsc-users mailing list<br>
> <a href="mailto:petsc-users@mcs.anl.gov">petsc-users@mcs.anl.gov</a><br>
> <a href="https://lists.mcs.anl.gov/mailman/listinfo/petsc-users" target="_blank">https://lists.mcs.anl.gov/mailman/listinfo/petsc-users</a><br>
><br>
><br>
> End of petsc-users Digest, Vol 43, Issue 61<br>
> *******************************************<br>
><br>
</blockquote></div><br>