Re: [Pw_forum] Questions about parallelization parameters
Thank you. This is very helpful On Sun, Aug 16, 2015 at 3:47 PM, Paolo Giannozzi wrote: > 1. Correct > 2. There are some hints (calling them "guidelines" is an overstatement) in > the user guide. For most cases, you need only to consider parallelization > on k-points (-nk), > on bands using "task groups" (-ntg), on linear algebra (-nd). > 3. The maximum number of processors for linear algebra parallelization is > the same as for PW parallelization,* before *subdivision into "task > groups". The latter are used only when computing V*psi_i and |psi_i|^2 > (instead of parallelizing over PW using all processors, one parallelizes > over PW *and* band index i) > > Paolo > > > On Sun, Aug 16, 2015 at 1:19 PM, mohammed shambakey > wrote: > >> Hi >> >> >>1. As I understood from user manual, the “world” of processors is >>devided into number of “images”. Each “image” is devided into number of >>“pools”. Each “pool” is devided into number of “bands”. “PW” level depends >>on the least specified one of the previous four (i.e., if “band” is >>specified, then “PW” is automatically distributed over all processor >>available in each “band”. If “pool” is the least specified parameter, then >>“PW” is automatically distributed over all processors available in each >>“pool”, and so on). Prcoessors of each “PW” is devided into number of >>“tasks” for 3D FFT. “Linear-algebra group” specifies number of processors >>that should not exceed number of processors availabel for each “PW”, >> right? >>2. Are there guidelines for the best values for the different >>parameters (i.e., ni, nk, nb, ... etc), or I should test each problem >>individually? >>3. If my understanding is correct, the user manual says “nd is set to >>1 if ScaLAPACK is not compiled, it is set to the square integer smaller >>than or equal to half the number of processors of each pool”. But “nd” >>should not exceed “PW”; and “PW” can be smaller than “half the number of >>processors of each pool” using “nt”. If this case happens, will there be >>error? >> >> Regards >> >> -- >> Mohammed >> >> ___ >> Pw_forum mailing list >> Pw_forum@pwscf.org >> http://pwscf.org/mailman/listinfo/pw_forum >> > > > > -- > Paolo Giannozzi, Dept. Chemistry&Physics&Environment, > Univ. Udine, via delle Scienze 208, 33100 Udine, Italy > Phone +39-0432-558216, fax +39-0432-558222 > > ___ > Pw_forum mailing list > Pw_forum@pwscf.org > http://pwscf.org/mailman/listinfo/pw_forum > -- Mohammed ___ Pw_forum mailing list Pw_forum@pwscf.org http://pwscf.org/mailman/listinfo/pw_forum
Re: [Pw_forum] Questions about parallelization parameters
1. Correct 2. There are some hints (calling them "guidelines" is an overstatement) in the user guide. For most cases, you need only to consider parallelization on k-points (-nk), on bands using "task groups" (-ntg), on linear algebra (-nd). 3. The maximum number of processors for linear algebra parallelization is the same as for PW parallelization,* before *subdivision into "task groups". The latter are used only when computing V*psi_i and |psi_i|^2 (instead of parallelizing over PW using all processors, one parallelizes over PW *and* band index i) Paolo On Sun, Aug 16, 2015 at 1:19 PM, mohammed shambakey wrote: > Hi > > >1. As I understood from user manual, the “world” of processors is >devided into number of “images”. Each “image” is devided into number of >“pools”. Each “pool” is devided into number of “bands”. “PW” level depends >on the least specified one of the previous four (i.e., if “band” is >specified, then “PW” is automatically distributed over all processor >available in each “band”. If “pool” is the least specified parameter, then >“PW” is automatically distributed over all processors available in each >“pool”, and so on). Prcoessors of each “PW” is devided into number of >“tasks” for 3D FFT. “Linear-algebra group” specifies number of processors >that should not exceed number of processors availabel for each “PW”, right? >2. Are there guidelines for the best values for the different >parameters (i.e., ni, nk, nb, ... etc), or I should test each problem >individually? >3. If my understanding is correct, the user manual says “nd is set to >1 if ScaLAPACK is not compiled, it is set to the square integer smaller >than or equal to half the number of processors of each pool”. But “nd” >should not exceed “PW”; and “PW” can be smaller than “half the number of >processors of each pool” using “nt”. If this case happens, will there be >error? > > Regards > > -- > Mohammed > > ___ > Pw_forum mailing list > Pw_forum@pwscf.org > http://pwscf.org/mailman/listinfo/pw_forum > -- Paolo Giannozzi, Dept. Chemistry&Physics&Environment, Univ. Udine, via delle Scienze 208, 33100 Udine, Italy Phone +39-0432-558216, fax +39-0432-558222 ___ Pw_forum mailing list Pw_forum@pwscf.org http://pwscf.org/mailman/listinfo/pw_forum
[Pw_forum] Questions about parallelization parameters
Hi 1. As I understood from user manual, the “world” of processors is devided into number of “images”. Each “image” is devided into number of “pools”. Each “pool” is devided into number of “bands”. “PW” level depends on the least specified one of the previous four (i.e., if “band” is specified, then “PW” is automatically distributed over all processor available in each “band”. If “pool” is the least specified parameter, then “PW” is automatically distributed over all processors available in each “pool”, and so on). Prcoessors of each “PW” is devided into number of “tasks” for 3D FFT. “Linear-algebra group” specifies number of processors that should not exceed number of processors availabel for each “PW”, right? 2. Are there guidelines for the best values for the different parameters (i.e., ni, nk, nb, ... etc), or I should test each problem individually? 3. If my understanding is correct, the user manual says “nd is set to 1 if ScaLAPACK is not compiled, it is set to the square integer smaller than or equal to half the number of processors of each pool”. But “nd” should not exceed “PW”; and “PW” can be smaller than “half the number of processors of each pool” using “nt”. If this case happens, will there be error? Regards -- Mohammed ___ Pw_forum mailing list Pw_forum@pwscf.org http://pwscf.org/mailman/listinfo/pw_forum