[petsc-users] PETSc-gamg default setting

[藤井昭宏]

Hello everybody,

I’d like to compare a solver with PETSc-gamg for Poisson problems.
I used PETSc version 3.4.3., and the following command line option.

-ksp_type “bcgs” -pc_type “gamg” -ksp_monitor -ksp_rtol 1.0E-7 -log_summary -pc_mg_log


Would you give me some information on the following questions?

- Does this option correspond to BICGSTAB solver with smoothed aggregation algebraic multigrid method?

- Are coarser level small distributed matrices re-distributed to reduce the parallelism, when the number of processes is very large?

- What kind of smoother will be used for this command line option?

- pc_mg_log option shows the MG Apply stage. Does this stage time correspond to the solver time except for multi-level setup time?
Does it include the time for BICGSTAB?


Thanks in advance.

Akihiro Fujii

[Matthew Knepley]

Post by 藤井昭宏
Hello everybody,
I’d like to compare a solver with PETSc-gamg for Poisson problems.
I used PETSc version 3.4.3., and the following command line option.
-ksp_type “bcgs” -pc_type “gamg” -ksp_monitor -ksp_rtol 1.0E-7
-log_summary -pc_mg_log
Would you give me some information on the following questions?
- Does this option correspond to BICGSTAB solver with smoothed aggregation
algebraic multigrid method?

Yes

Post by 藤井昭宏
- Are coarser level small distributed matrices re-distributed to reduce
the parallelism, when the number of processes is very large?

Yes

Post by 藤井昭宏
- What kind of smoother will be used for this command line option?

Cheby(2)/Jacobi is the default in GAMG.

Post by 藤井昭宏
- pc_mg_log option shows the MG Apply stage. Does this stage time
correspond to the solver time except for multi-level setup time?
Does it include the time for BICGSTAB?

Yes, and no.

Thanks,

Matt

Post by 藤井昭宏
Thanks in advance.
Akihiro Fujii

[Barry Smith]

Run with -ksp_view (or -snes_view etc). If that does not provide all relevant information then tell us what is missing so we can add that to the output. Defaults etc change over time so you should never rely on what someone thinks it is or what it “should be”, just run with the viewer.

barry

Post by 藤井昭宏
Hello everybody,
I’d like to compare a solver with PETSc-gamg for Poisson problems.
I used PETSc version 3.4.3., and the following command line option.
-ksp_type “bcgs” -pc_type “gamg” -ksp_monitor -ksp_rtol 1.0E-7 -log_summary -pc_mg_log
Would you give me some information on the following questions?
- Does this option correspond to BICGSTAB solver with smoothed aggregation algebraic multigrid method?
Yes
- Are coarser level small distributed matrices re-distributed to reduce the parallelism, when the number of processes is very large?
Yes
- What kind of smoother will be used for this command line option?
Cheby(2)/Jacobi is the default in GAMG.
- pc_mg_log option shows the MG Apply stage. Does this stage time correspond to the solver time except for multi-level setup time?
Does it include the time for BICGSTAB?
Yes, and no.
Thanks,
Matt
Thanks in advance.
Akihiro Fujii
--
What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
-- Norbert Wiener

[藤井昭宏]

Thanks for your quick reply!
It helps me much.

Akihiro

Post by Barry Smith
Run with -ksp_view (or -snes_view etc). If that does not provide all relevant information then tell us what is missing so we can add that to the output. Defaults etc change over time so you should never rely on what someone thinks it is or what it “should be”, just run with the viewer.
barry

Post by 藤井昭宏
Hello everybody,
I’d like to compare a solver with PETSc-gamg for Poisson problems.
I used PETSc version 3.4.3., and the following command line option.
-ksp_type “bcgs” -pc_type “gamg” -ksp_monitor -ksp_rtol 1.0E-7 -log_summary -pc_mg_log
Would you give me some information on the following questions?
- Does this option correspond to BICGSTAB solver with smoothed aggregation algebraic multigrid method?
Yes
- Are coarser level small distributed matrices re-distributed to reduce the parallelism, when the number of processes is very large?
Yes
- What kind of smoother will be used for this command line option?
Cheby(2)/Jacobi is the default in GAMG.
- pc_mg_log option shows the MG Apply stage. Does this stage time correspond to the solver time except for multi-level setup time?
Does it include the time for BICGSTAB?
Yes, and no.
Thanks,
Matt
Thanks in advance.
Akihiro Fujii
--
What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
-- Norbert Wiener