[Pw_forum] How to choose number of pools
In data 30 aprile 2009 alle ore 16:57:46, Ihsan Erikat ha scritto: > Thank you, the CPU are mpd connection I don't know what it is, if it's not faster that a 10/100/1000 ethernet card your best bet is to use 8 processors and 8 pools, leaving 2 processors idle. Also, make sure that your outdir is set to some directory local to the nodes, e.g. /tmp, otherwise the overhead introduced when saving the restart data may kill the performance. best regards -- Lorenzo Paulatto SISSA & DEMOCRITOS (Trieste) phone: +39 040 3787 511 skype: paulatz www: http://people.sissa.it/~paulatto/ *** save italian brains *** http://saveitalianbrains.wordpress.com/
[Pw_forum] phonon calculation for large system
On Apr 30, 2009, at 10:13 , Gabriele Sclauzero wrote: > I think you can use this parallelization level even on a cluster, > so that you do not have to do modes and irreps sequentially, > but you can split them in concurrently running (multiprocessor) > jobs so that you-ll have an actual speed-up. sure, but it is currently not implemented in a MPI framework: it has to be done "by hand". It isn't difficult to implement it, using the "image group", though. > Or are you forced to use a single processor in order to exploit > this start_q, last_q,... keywords? no, I don't think so, or if so, it can be easily fixed Paolo --- Paolo Giannozzi, Democritos and University of Udine, Italy
[Pw_forum] How to choose number of pools
In data 30 aprile 2009 alle ore 16:18:38, Ihsan Erikat ha scritto: > Hello all > I am? running pwscf on 10 CPU how are they connected to each other? > and? the k points in my file is 3x6x1. how many kpoints do you have after symmetry operations have been applied? > When I use? the command mpirun -np xx? pw.x -npool x , how can I choose? > the effective number of processors and pools. All processors inside the same pool must be able to comunicate with each other with an extremely large bandwidth and low latency. You can have a benefit by taking the number of pools as a divisor of the number of kpoints, especially if the number of kpoints is small. regards -- Lorenzo Paulatto SISSA & DEMOCRITOS (Trieste) phone: +39 040 3787 511 skype: paulatz www: http://people.sissa.it/~paulatto/ *** save italian brains *** http://saveitalianbrains.wordpress.com/
[Pw_forum] transmission calculation
Dear Manoj
Please provide also scf files for the leads.
Note that in the case ikind=2 (different leads) the scattering file
should be a double unit cell i.e. should contain the scattering barrier
+ some portion of the right lead + the scattering barrier in order to
have regular periodicity in the z direction. You should also provide
bds parameter which says where the scattering barrier ends (while it
should start from 0).
Regards, Alexander
On Thu, 2009-04-30 at 07:23 -0400, Manoj Srivastava wrote:
> Dear PWSCF users and developers,
> I have tried to do a transmission calcuation and not getting right
> results. I have tried to figure out the reason behind this and realized
> that in my set up for some reason the Bloch waves in left lead get changed
> resulting wrong transmission. By this I meant that, if I just try to
> calculate complex band of the left lead, I have a kz for each kx,ky and E
> but when I am doing transmission calculation, the Bloch's wave in the same
> lead has different kz for the same kx,ky and E, which does not make sense
> as why should the Bloch's wave in lead get affected by the presence of
> scattering region? For the right lead kz remains the same in both cases
> though, which I believe how it should be.
> I did this for several cases, I interchanged right and left leads,
> increased number of atoms in the scattering region, chose scattering
> region in different way, and everytime kz of left lead changes. Following
> is my input file for the scattering region -
>
> scf part -
>
>
> calculation='scf'
> pseudo_dir = '/home/manoj/espresso-4.0.1/pseudo',
> outdir='./',
> prefix='cus',
> /
>
> ibrav =0,
> celldm(1)=4.82,
> nat= 10,
> ntyp= 1,
> ecutwfc =50.0,
> occupations='smearing',
> smearing='gaussian',
> degauss=0.02,
> ecutrho=500
> /
>
> conv_thr = 1.0e-8
> mixing_beta=0.7
> /
> ATOMIC_SPECIES
> Cu 63.55 Cu.pz-d-rrkjus.UPF
> ATOMIC_POSITIONS
> Cu 0.0 0.0 0.0
> Cu -0.5 0.288675134 0.816496581
> Cu -1.0 0.577350269 1.632993162
> Cu -1.5 0.866025404 2.449489743
> Cu 0.0 0.0 3.265986324
> Cu 0.5 -0.288675134 4.082482905
> Cu 1.0 -0.577350269 4.898979486
> Cu 1.5 -0.866025404 5.715476066
> Cu 1.0 -0.577350269 6.531972647
> Cu 0.5 -0.288675134 7.348469228
> K_POINTS (automatic)
> 6 6 4 1 1 1
> CELL_PARAMETERS {hexagonal}
> 1.0 0.0 0.0
> -0.5 0.866025403 0.0
> 0.0 0.0 8.16496581
>
>
>
> conductance part -
>
>
> outdir='./'
> prefixl='cul'
> prefixr='cur'
> prefixs='cus'
> tran_file ='trans.twin'
> ikind=2
> ecut2d=50
> energy0=0.0
> denergy=0.0
> ewind=5.d0
> epsproj=1.d-6
> nz1=25
> /
> 2
> 0.00 0.00 1
> 0.500 -0.500 1
>
> 1
> 0.0
>
> Thank you for your attention.
>
> Regards,
> Manoj Srivastava
> Graduate Student
> Department of Physics
> University of Florida, Gainesville, FL
>
> ___
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
--
Alexander Smogunov
Email: smogunov at sissa.it
Home page: http://people.sissa.it/~smogunov
[Pw_forum] transmission calculation
Manoj Srivastava wrote:
> Dear PWSCF users and developers,
> I have tried to do a transmission calcuation and not getting right
> results. I have tried to figure out the reason behind this and realized
> that in my set up for some reason the Bloch waves in left lead get changed
> resulting wrong transmission. By this I meant that, if I just try to
> calculate complex band of the left lead, I have a kz for each kx,ky and E
> but when I am doing transmission calculation, the Bloch's wave in the same
> lead has different kz for the same kx,ky and E, which does not make sense
> as why should the Bloch's wave in lead get affected by the presence of
> scattering region? For the right lead kz remains the same in both cases
> though, which I believe how it should be.
Does it happen also when you do not use G_perp basis reduction? Can you try
again the CBS
and the transmission calculation without specifying ecut2d, nz1 and putting
ewind=1.d3?
Better, first check the size of the reduced basis using
grep n2d
and see if they are very different (or if the second is too low with respect to
the size
of the whole basis set).
Also try to understand which of the two results (kz vectors) is correct, by
comparing the
real part of kz with the band structure from pw.x
GS
> I did this for several cases, I interchanged right and left leads,
> increased number of atoms in the scattering region, chose scattering
> region in different way, and everytime kz of left lead changes. Following
> is my input file for the scattering region -
>
> scf part -
>
>
> calculation='scf'
> pseudo_dir = '/home/manoj/espresso-4.0.1/pseudo',
> outdir='./',
> prefix='cus',
> /
>
> ibrav =0,
> celldm(1)=4.82,
> nat= 10,
> ntyp= 1,
> ecutwfc =50.0,
> occupations='smearing',
> smearing='gaussian',
> degauss=0.02,
> ecutrho=500
> /
>
> conv_thr = 1.0e-8
> mixing_beta=0.7
> /
> ATOMIC_SPECIES
> Cu 63.55 Cu.pz-d-rrkjus.UPF
> ATOMIC_POSITIONS
> Cu 0.0 0.0 0.0
> Cu -0.5 0.288675134 0.816496581
> Cu -1.0 0.577350269 1.632993162
> Cu -1.5 0.866025404 2.449489743
> Cu 0.0 0.0 3.265986324
> Cu 0.5 -0.288675134 4.082482905
> Cu 1.0 -0.577350269 4.898979486
> Cu 1.5 -0.866025404 5.715476066
> Cu 1.0 -0.577350269 6.531972647
> Cu 0.5 -0.288675134 7.348469228
> K_POINTS (automatic)
> 6 6 4 1 1 1
> CELL_PARAMETERS {hexagonal}
> 1.0 0.0 0.0
> -0.5 0.866025403 0.0
> 0.0 0.0 8.16496581
>
>
>
> conductance part -
>
>
> outdir='./'
> prefixl='cul'
> prefixr='cur'
> prefixs='cus'
> tran_file ='trans.twin'
> ikind=2
> ecut2d=50
> energy0=0.0
> denergy=0.0
> ewind=5.d0
> epsproj=1.d-6
> nz1=25
> /
> 2
> 0.00 0.00 1
> 0.500 -0.500 1
>
> 1
> 0.0
>
> Thank you for your attention.
>
> Regards,
> Manoj Srivastava
> Graduate Student
> Department of Physics
> University of Florida, Gainesville, FL
>
> ___
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
--
o o
| Gabriele Sclauzero, PhD Student |
| c/o: SISSA & CNR-INFM Democritos, |
|via Beirut 2-4, 34014 Trieste (Italy) |
| email: sclauzer at sissa.it |
| phone: +39 040 3787 511 |
| skype: gurlonotturno |
o o
[Pw_forum] transmission calculation
In data 30 aprile 2009 alle ore 13:23:25, Manoj Srivastava ha scritto: > I did this for several cases, I interchanged right and left leads, > increased number of atoms in the scattering region, chose scattering > region in different way, and everytime kz of left lead changes. Following > is my input file for the scattering region - I'm no expert in transimission calculation, but I have had a look at your scf input file for the lead region, using XCrysDen, and I've noticed tha tlayers 4 and 5 are aligned in a strange way, reducing the system symmetry. Are you sure you have done it on purpose? Have you tried checking your input files with XCD? best regards -- Lorenzo Paulatto SISSA & DEMOCRITOS (Trieste) phone: +39 040 3787 511 skype: paulatz www: http://people.sissa.it/~paulatto/ *** save italian brains *** http://saveitalianbrains.wordpress.com/
[Pw_forum] phonon calculation for large system
Huiqun Zhou wrote: > Gabriele, > > I just downloaded the CVS version two days ago, but found no INPUT_PH > in Doc. Should you please post one, or let me know how to generate it? try: cvs update -d doc-def you should find then a directory doc-def inside your cvs tree, and inside there will be template files for automatic generation of the documentation. From inside doc-def, type make but this probably won't work unless you have all needed tcl libraries installed. Anyway you can read directly the INPUT_PH.def file, searching for start_q, start_irr, etc. etc. GS > > Huiqun Zhou > @Earth Sciences, Nanjing University, China > > > - Original Message - > From: "Gabriele Sclauzero" > To: "PWSCF Forum" > Sent: Thursday, April 30, 2009 3:46 PM > Subject: Re: [Pw_forum] phonon calculation for large system > > >> >> zq wu wrote: >>> Dear pwscf users, >>> >>> I need to do phonon calculation for large system (several hundreds atoms >>> per cell). Since i can use lots of processors, I plan to submit many >>> jobs, each of which will calculate 1 phonon mode. But i do not know >>> whether it is available to get final dynamic matrix from these separate >>> runs and how to do it. Can any one help me? >>> >>> Currently, phonon modes are calculated one by one in sequence in phonon >>> calculation. Can we do some parallelization here? Can we divide the >>> processors into several groups and let each group take care of one or >>> more modes in a way similar to K-points parallelization? I think that >>> the parallel efficeincy for modes parallelization should also be close >>> to K-points parallelization since the phonon modes calculation is >>> independent of each other. The modes parallelization should be very >>> useful for system with large numbers of atoms. Does anyone have the >>> idea how hard it is to do the modes parallelization? >> If I correctly understood, i think that this kind of parallelization has >> already been >> exploited in the latest versions of CVS, in order to use ph.x on the GRID. >> I don't know if >> all the related problems have been fixed at this stage (have they, >> Paolo?), but you can >> try to download the CVS and use the keywords start_q, last_q, start_irr, >> last_irr (see >> INPUT_PH) to split your job in many independent jobs. Sorry, I don't know >> the details, but >> at least now you're aware that there is this possibility available (though >> not extensively >> tested, I think, so use _with_ _care_ and help debugging, please). >> >> HTH >> >> GS >> >> >>> Thanks >>> >>> Zhongqing >>> >>> CACS University of Southern California >>> >>> >>> >>> >>> >>> ___ >>> Pw_forum mailing list >>> Pw_forum at pwscf.org >>> http://www.democritos.it/mailman/listinfo/pw_forum >> -- >> >> >> o o >> | Gabriele Sclauzero, PhD Student | >> | c/o: SISSA & CNR-INFM Democritos, | >> |via Beirut 2-4, 34014 Trieste (Italy) | >> | email: sclauzer at sissa.it | >> | phone: +39 040 3787 511 | >> | skype: gurlonotturno | >> o o >> ___ >> Pw_forum mailing list >> Pw_forum at pwscf.org >> http://www.democritos.it/mailman/listinfo/pw_forum >> > > ___ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum > -- o o | Gabriele Sclauzero, PhD Student | | c/o: SISSA & CNR-INFM Democritos, | |via Beirut 2-4, 34014 Trieste (Italy) | | email: sclauzer at sissa.it | | phone: +39 040 3787 511 | | skype: gurlonotturno | o o
[Pw_forum] transmission calculation
Dear Gabriele,
Thanks for answering.
On Thu, 30 Apr 2009, Gabriele Sclauzero wrote:
> Manoj Srivastava wrote:
> > Dear PWSCF users and developers,
> > I have tried to do a transmission calcuation and not getting right
> > results. I have tried to figure out the reason behind this and realized
> > that in my set up for some reason the Bloch waves in left lead get changed
> > resulting wrong transmission. By this I meant that, if I just try to
> > calculate complex band of the left lead, I have a kz for each kx,ky and E
> > but when I am doing transmission calculation, the Bloch's wave in the same
> > lead has different kz for the same kx,ky and E, which does not make sense
> > as why should the Bloch's wave in lead get affected by the presence of
> > scattering region? For the right lead kz remains the same in both cases
> > though, which I believe how it should be.
>
> Does it happen also when you do not use G_perp basis reduction? Can you try
> again the CBS
> and the transmission calculation without specifying ecut2d, nz1 and putting
> ewind=1.d3?
Yes, I have tried this. No difference, the kz value for left lead gets
changed, when i switch from ikind=0 to ikind=2.
>
> Better, first check the size of the reduced basis using
>
> grep n2d
>
> and see if they are very different (or if the second is too low with respect
> to the size
> of the whole basis set).
> Also try to understand which of the two results (kz vectors) is correct, by
> comparing the
> real part of kz with the band structure from pw.x
The one with ikind=0 is correct.
>
>
> GS
>
>
> > I did this for several cases, I interchanged right and left leads,
> > increased number of atoms in the scattering region, chose scattering
> > region in different way, and everytime kz of left lead changes. Following
> > is my input file for the scattering region -
> >
> > scf part -
> >
> >
> > calculation='scf'
> > pseudo_dir = '/home/manoj/espresso-4.0.1/pseudo',
> > outdir='./',
> > prefix='cus',
> > /
> >
> > ibrav =0,
> > celldm(1)=4.82,
> > nat= 10,
> > ntyp= 1,
> > ecutwfc =50.0,
> > occupations='smearing',
> > smearing='gaussian',
> > degauss=0.02,
> > ecutrho=500
> > /
> >
> > conv_thr = 1.0e-8
> > mixing_beta=0.7
> > /
> > ATOMIC_SPECIES
> > Cu 63.55 Cu.pz-d-rrkjus.UPF
> > ATOMIC_POSITIONS
> > Cu 0.0 0.0 0.0
> > Cu -0.5 0.288675134 0.816496581
> > Cu -1.0 0.577350269 1.632993162
> > Cu -1.5 0.866025404 2.449489743
> > Cu 0.0 0.0 3.265986324
> > Cu 0.5 -0.288675134 4.082482905
> > Cu 1.0 -0.577350269 4.898979486
> > Cu 1.5 -0.866025404 5.715476066
> > Cu 1.0 -0.577350269 6.531972647
> > Cu 0.5 -0.288675134 7.348469228
> > K_POINTS (automatic)
> > 6 6 4 1 1 1
> > CELL_PARAMETERS {hexagonal}
> > 1.0 0.0 0.0
> > -0.5 0.866025403 0.0
> > 0.0 0.0 8.16496581
> >
> >
> >
> > conductance part -
> >
> >
> > outdir='./'
> > prefixl='cul'
> > prefixr='cur'
> > prefixs='cus'
> > tran_file ='trans.twin'
> > ikind=2
> > ecut2d=50
> > energy0=0.0
> > denergy=0.0
> > ewind=5.d0
> > epsproj=1.d-6
> > nz1=25
> > /
> > 2
> > 0.00 0.00 1
> > 0.500 -0.500 1
> >
> > 1
> > 0.0
> >
> > Thank you for your attention.
> >
> > Regards,
> > Manoj Srivastava
> > Graduate Student
> > Department of Physics
> > University of Florida, Gainesville, FL
> >
> > ___
> > Pw_forum mailing list
> > Pw_forum at pwscf.org
> > http://www.democritos.it/mailman/listinfo/pw_forum
> >
>
> --
>
>
> o o
> | Gabriele Sclauzero, PhD Student |
> | c/o: SISSA & CNR-INFM Democritos, |
> |via Beirut 2-4, 34014 Trieste (Italy) |
> | email: sclauzer at sissa.it |
> | phone: +39 040 3787 511 |
> | skype: gurlonotturno |
> o o
> ___
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>
[Pw_forum] transmission calculation
Dear Lorenzo, Thank you for answering. Layers 4 and 5 are supposed be like that, it is done on purpose. The file I provided was not for the lead but for the scattering region. I have visualized my structure, it looks ok. Regards, Manoj On Thu, 30 Apr 2009, Lorenzo Paulatto wrote: > In data 30 aprile 2009 alle ore 13:23:25, Manoj Srivastava > ha scritto: > > I did this for several cases, I interchanged right and left leads, > > increased number of atoms in the scattering region, chose scattering > > region in different way, and everytime kz of left lead changes. Following > > is my input file for the scattering region - > > I'm no expert in transimission calculation, but I have had a look at your > scf input file for the lead region, using XCrysDen, and I've noticed tha > tlayers 4 and 5 are aligned in a strange way, reducing the system > symmetry. Are you sure you have done it on purpose? Have you tried > checking your input files with XCD? > > best regards > > > -- > Lorenzo Paulatto > SISSA & DEMOCRITOS (Trieste) > phone: +39 040 3787 511 > skype: paulatz > www: http://people.sissa.it/~paulatto/ > > *** save italian brains *** > http://saveitalianbrains.wordpress.com/ > ___ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum >
[Pw_forum] electron-phonon coupling within rigid band approximation
> I guess you can, but is this what you want? Yes, this would be equivalent calculating electron-phonon coupling of a doped material, and this is what I actually need. (In the hope that rigid-band approximation is ok for this system at small doping levels). I assume I need to do the following steps. First, I do the scf calculation for the insulator with very small gaussian smearing. Then modify the elphon.f90 file to add/substract a number to effit(isig) after the call to efermig. Please let me know if there is a better way. Thanks, Alaska
[Pw_forum] How to choose number of pools
On Thu, 2009-04-30 at 17:09 +0200, Lorenzo Paulatto wrote: > In data 30 aprile 2009 alle ore 16:57:46, Ihsan Erikat > > ha scritto: > > > Thank you, the CPU are mpd connection lorenzo, mpd are the processes that MPICH-2 generates upon initializing the parallel environment. that doesn't really help much, but it is likely that the cluster is running with GigE at best. > I don't know what it is, if it's not faster that a 10/100/1000 ethernet > card your best bet is to use 8 processors and 8 pools, leaving 2 > processors idle. Also, make sure that your outdir is set to some directory better yet. run with -np 8 and put a second job on the other two. ;-) cheers, axel. > local to the nodes, e.g. /tmp, otherwise the overhead introduced when > saving the restart data may kill the performance. > > best regards > > > > > -- === Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu Center for Molecular Modeling -- University of Pennsylvania Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323 tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425 === If you make something idiot-proof, the universe creates a better idiot.
[Pw_forum] phonon calculation for large system
On Thu, Apr 30, 2009 at 7:28 AM, Paolo Giannozzi wrote: > > On Apr 30, 2009, at 10:13 , Gabriele Sclauzero wrote: > > > I think you can use this parallelization level even on a cluster, > > so that you do not have to do modes and irreps sequentially, > > but you can split them in concurrently running (multiprocessor) > > jobs so that you-ll have an actual speed-up. > > sure, but it is currently not implemented in a MPI framework: > it has to be done "by hand". It isn't difficult to implement it, > using the "image group", though. > > > Or are you forced to use a single processor in order to exploit > > this start_q, last_q,... keywords? > > no, I don't think so, or if so, it can be easily fixed > > Paolo > --- > Paolo Giannozzi, Democritos and University of Udine, Italy > > Can single-q, single-mode calculation run in several nodes. If so, memory may be not a problem even for system with several hundreds atoms. Thank you very much. All these information are very helpful. I will try the latest versions of CVS. Zhongqing > > ___ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum > -- next part -- An HTML attachment was scrubbed... URL: http://www.democritos.it/pipermail/pw_forum/attachments/20090430/5ed0adcd/attachment.htm
[Pw_forum] How to choose number of pools
On Thu, 2009-04-30 at 07:18 -0700, Ihsan Erikat wrote: please check out previous lengthy discussions on the subject in the mailing list archives _again_. this topic has been discussed repeatedly and people have done benchmarks to demonstrate and understand how the various parallelization options help under specific environments. particularly the number of cpus per node, cores per cpu, main memory per cpu/core, cache memory per cpu/core in combination with the interconnect (tcp/ip, infiniband/myrinet/...) can lead to different results. in some cases it may even better to not use all cpu cores in order maximize cache usage, in others it may be important to use the more communication demanding parallelization scheme to reduce the per process memory use. ... and this is only the beginning. compiler vendor and flags as well as BLAS/LAPACK and communication libraries can have and impact, not to mention the setup and configuration of machines. (e.g. modern linux distributions come with SELinux enabled which effectively trashes the cache by randomizing the stack frame location and lots of useless auto-configuration demons running that interrupt the kernel needlessly and thus create a lot of OS jitter which affects latency sensitive applications. running a "lean mean computing machine" requires some effort on many levels...). in short: do tests and figure it out yourself. what works best depends very much on your input set size, network, hard- and software. cheers, axel. > Hello all > I am running pwscf on 10 CPU and the k points in my file is 3x6x1. > > When I use the command mpirun -np xx pw.x -npool x , how can I > choose the effective number of processors and pools . I read the > manual and the forum archive, but I don`t know how to choose these > numbers. I`ll appreciate any help. > > > > best regards > Ihsan Erikat > Physics department-Jordan University > > > > ___ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum -- === Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu Center for Molecular Modeling -- University of Pennsylvania Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323 tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425 === If you make something idiot-proof, the universe creates a better idiot.
[Pw_forum] phonon calculation for large system
Paolo Giannozzi wrote: > On Apr 30, 2009, at 9:46 , Gabriele Sclauzero wrote: > >> i think that this kind of parallelization has already been exploited >> in the latest versions of CVS, in order to use ph.x on the GRID. >> I don't know if all the related problems have been fixed at this stage > > they have, but I am not sure you can run hundreds of atoms on the grid: > the single-q, single-mode calculation has to fit into the memory of > each node. Of course, you are right. Anyway, I think you can use this parallelization level even on a cluster, so that you do not have to do modes and irreps sequentially, but you can split them in concurrently running (multiprocessor) jobs so that you-ll have an actual speed-up. Or are you forced to use a single processor in order to exploit this start_q, last_q,... keywords? GS > > P. > --- > Paolo Giannozzi, Democritos and University of Udine, Italy > > > ___ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum > -- o o | Gabriele Sclauzero, PhD Student | | c/o: SISSA & CNR-INFM Democritos, | |via Beirut 2-4, 34014 Trieste (Italy) | | email: sclauzer at sissa.it | | phone: +39 040 3787 511 | | skype: gurlonotturno | o o
[Pw_forum] phonon calculation for large system
On Apr 30, 2009, at 9:46 , Gabriele Sclauzero wrote: > i think that this kind of parallelization has already been exploited > in the latest versions of CVS, in order to use ph.x on the GRID. > I don't know if all the related problems have been fixed at this stage they have, but I am not sure you can run hundreds of atoms on the grid: the single-q, single-mode calculation has to fit into the memory of each node. P. --- Paolo Giannozzi, Democritos and University of Udine, Italy
[Pw_forum] phonon calculation for large system
zq wu wrote: > Dear pwscf users, > > I need to do phonon calculation for large system (several hundreds atoms > per cell). Since i can use lots of processors, I plan to submit many > jobs, each of which will calculate 1 phonon mode. But i do not know > whether it is available to get final dynamic matrix from these separate > runs and how to do it. Can any one help me? > > Currently, phonon modes are calculated one by one in sequence in phonon > calculation. Can we do some parallelization here? Can we divide the > processors into several groups and let each group take care of one or > more modes in a way similar to K-points parallelization? I think that > the parallel efficeincy for modes parallelization should also be close > to K-points parallelization since the phonon modes calculation is > independent of each other. The modes parallelization should be very > useful for system with large numbers of atoms. Does anyone have the > idea how hard it is to do the modes parallelization? If I correctly understood, i think that this kind of parallelization has already been exploited in the latest versions of CVS, in order to use ph.x on the GRID. I don't know if all the related problems have been fixed at this stage (have they, Paolo?), but you can try to download the CVS and use the keywords start_q, last_q, start_irr, last_irr (see INPUT_PH) to split your job in many independent jobs. Sorry, I don't know the details, but at least now you're aware that there is this possibility available (though not extensively tested, I think, so use _with_ _care_ and help debugging, please). HTH GS > > Thanks > > Zhongqing > > CACS University of Southern California > > > > > > ___ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum -- o o | Gabriele Sclauzero, PhD Student | | c/o: SISSA & CNR-INFM Democritos, | |via Beirut 2-4, 34014 Trieste (Italy) | | email: sclauzer at sissa.it | | phone: +39 040 3787 511 | | skype: gurlonotturno | o o
[Pw_forum] electron-phonon coupling within rigid band approximation
On Apr 30, 2009, at 4:58 , asubedi wrote: > I am trying to calculate electron-phonon coupling of an insulator, > which is not possible with QE. However, can I do it by fixing the > Fermi energy so that it crosses the valence bands? I guess you can, but is this what you want? Paolo --- Paolo Giannozzi, Democritos and University of Udine, Italy -- next part -- An HTML attachment was scrubbed... URL: http://www.democritos.it/pipermail/pw_forum/attachments/20090430/a4593525/attachment.htm
[Pw_forum] problem with installing Quantum ESPRESSO CVS version
On Apr 29, 2009, at 23:43 , shruba at gmail.com wrote: > /apps/intel/fce/10.1.008/lib/libimf.so: warning: warning: > feupdateenv is not implemented > and will always fail not a problem > /opt/intel/mkl/10.0.1.014/lib/em64t/libmkl_lapack.so: undefined > reference to `mkl_serv_load_fun' in your make.sys, you should have LDFLAGS = -i-static -openmp BLAS_LIBS = -L /opt/intel/mkl/10.0.1.014/lib/em64t -lmkl_em64t LAPACK_LIBS= and no -lmkl_lapack -lguide -lpthreads > when I contacted to the cluster management they suggested me to check > weather CVS version is compatible with version 10.0.1.014 of the > intel mkl? of course it is and it works on all machines I have tried. If you find why it doesn't in your case, please report. Most likely, the version of the intel compiler found by configure is either old (<v.9) or not correctly recognized for some obscure reason. Paolo --- Paolo Giannozzi, Democritos and University of Udine, Italy -- next part -- An HTML attachment was scrubbed... URL: http://www.democritos.it/pipermail/pw_forum/attachments/20090430/ec2611ad/attachment.htm
[Pw_forum] How to choose number of pools
Thank you, the CPU are mpd connection and the home is NFS share. When I run the job in one CPU the output file gives k-point =8, best regards Ihsan Erikat Physics department , Jordan University -- next part -- An HTML attachment was scrubbed... URL: http://www.democritos.it/pipermail/pw_forum/attachments/20090430/03682ba3/attachment.htm
[Pw_forum] transmission calculation
Dear PWSCF users and developers,
I have tried to do a transmission calcuation and not getting right
results. I have tried to figure out the reason behind this and realized
that in my set up for some reason the Bloch waves in left lead get changed
resulting wrong transmission. By this I meant that, if I just try to
calculate complex band of the left lead, I have a kz for each kx,ky and E
but when I am doing transmission calculation, the Bloch's wave in the same
lead has different kz for the same kx,ky and E, which does not make sense
as why should the Bloch's wave in lead get affected by the presence of
scattering region? For the right lead kz remains the same in both cases
though, which I believe how it should be.
I did this for several cases, I interchanged right and left leads,
increased number of atoms in the scattering region, chose scattering
region in different way, and everytime kz of left lead changes. Following
is my input file for the scattering region -
scf part -
calculation='scf'
pseudo_dir = '/home/manoj/espresso-4.0.1/pseudo',
outdir='./',
prefix='cus',
/
ibrav =0,
celldm(1)=4.82,
nat= 10,
ntyp= 1,
ecutwfc =50.0,
occupations='smearing',
smearing='gaussian',
degauss=0.02,
ecutrho=500
/
conv_thr = 1.0e-8
mixing_beta=0.7
/
ATOMIC_SPECIES
Cu 63.55 Cu.pz-d-rrkjus.UPF
ATOMIC_POSITIONS
Cu 0.0 0.0 0.0
Cu -0.5 0.288675134 0.816496581
Cu -1.0 0.577350269 1.632993162
Cu -1.5 0.866025404 2.449489743
Cu 0.0 0.0 3.265986324
Cu 0.5 -0.288675134 4.082482905
Cu 1.0 -0.577350269 4.898979486
Cu 1.5 -0.866025404 5.715476066
Cu 1.0 -0.577350269 6.531972647
Cu 0.5 -0.288675134 7.348469228
K_POINTS (automatic)
6 6 4 1 1 1
CELL_PARAMETERS {hexagonal}
1.0 0.0 0.0
-0.5 0.866025403 0.0
0.0 0.0 8.16496581
conductance part -
outdir='./'
prefixl='cul'
prefixr='cur'
prefixs='cus'
tran_file ='trans.twin'
ikind=2
ecut2d=50
energy0=0.0
denergy=0.0
ewind=5.d0
epsproj=1.d-6
nz1=25
/
2
0.00 0.00 1
0.500 -0.500 1
1
0.0
Thank you for your attention.
Regards,
Manoj Srivastava
Graduate Student
Department of Physics
University of Florida, Gainesville, FL
[Pw_forum] How to choose number of pools
Hello all I am? running pwscf on 10 CPU and? the k points in my file is 3x6x1. ? When I use? the command mpirun -np xx? pw.x -npool x , how can I choose? the effective number of processors and pools . I read the manual and the forum archive, but I don`t know how to choose these numbers. I`ll appreciate any help. best regards Ihsan Erikat Physics department-Jordan University -- next part -- An HTML attachment was scrubbed... URL: http://www.democritos.it/pipermail/pw_forum/attachments/20090430/d23ac12e/attachment.htm
