GW parallel strategies

From The Yambo Project
Jump to navigation Jump to search

This modules contains very general discussions of the parallel environment of Yambo. In this tutorial we will see how to setup the variables governing the parallel execution of yambo in order to perform efficient calculations in terms of both cpu time and memory to solution. As a test case we will consider the hBN 2D material. Because of its reduced dimensionality, GW calculations turns out to be very delicate. Beside the usual convergence studies with respect to k-points and sums-over-bands, in low dimensional systems a sensible amount of vacuum is required in order to treat the system as isolated, translating into a large number of plane-waves. As for other tutorials, it is important to stress that this tutorial it is meant to illustrate the functionality of the key variables and to run in reasonable time, so it has not the purpose to reach the desired accuracy to reproduce experimental results. Moreover please also note that scaling performance illustrated below may be significantly dependent on the underlying parallel architecture. Nevertheless, general considerations are tentatively drawn in discussing the results.

Getting familiar with yambo in parallel

Let's start by copying the tutorial files in the cluster and unzip them in the folder you will run the tutorial.

$ cp $path/hBN-2D-para.tar.gz ./
$ tar -zxvf hBN-2D-para.tar.gz
$ cd ./hBN-2D-para/YAMBO

Under the YAMBO folder, together with the SAVE folder, you will see the script

$ ls  SAVE

First, run the initialization as usual. Then you need to generate the input file for a GW run.

$ yambo -g n -p p -F 

After setting the variables in red, the new input file should look like the following:

$ cat
ppa                          # [R Xp] Plasmon Pole Approximation
gw0                          # [R GW] GoWo Quasiparticle energy levels
HF_and_locXC                 # [R XX] Hartree-Fock Self-energy and Vxc
em1d                         # [R Xd] Dynamical Inverse Dielectric Matrix
X_Threads=  0                # [OPENMP/X] Number of threads for response functions 
DIP_Threads=  0              # [OPENMP/X] Number of threads for dipoles
SE_Threads=  0               # [OPENMP/GW] Number of threads for self-energy
EXXRLvcs= 21817        RL    # [XX] Exchange RL components
VXCRLvcs= 21817        RL    # [XC] XCpotential RL components
Chimod= ""                   # [X] IP/Hartree/ALDA/LRC/BSfxc
% BndsRnXp
    1 |  300 |               # [Xp] Polarization function bands
NGsBlkXp= 4            Ry    # [Xp] Response block size
% LongDrXp
 1.000000 | 0.000000 | 0.000000 |        # [Xp] [cc] Electric Field
PPAPntXp= 27.21138     eV    # [Xp] PPA imaginary energy
% GbndRnge
    1 |  300 |               # [GW] G[W] bands range
GDamping=  0.10000     eV    # [GW] G[W] damping
dScStep=  0.10000      eV    # [GW] Energy step to evaluate Z factors
DysSolver= "n"               # [GW] Dyson Equation solver ("n","s","g")
%QPkrange                    # [GW] QP generalized Kpoint/Band indices
  1| 4| 1| 8|

Now you need to create a submission script. here below an example ( based on the SLURM scheduler. In the case of other schedulers, the header should be updated accordingly.

$ cat
#SBATCH -t 06:00:00 
#SBATCH -J test
#SBATCH --partition=<queue name>
#SBATCH --tasks-per-node=1
#SBATCH --cpus-per-task=1

ncpu=`echo $nodes $tasks_per_node | awk '{print $1*$2}'`

module purge
module load <needed modules> 
module load <more modules> 
bindir=<path to yambo bindir> 

export OMP_NUM_THREADS=$nthreads


cp -f $filein0 $filein
cat >> $filein << EOF

DIP_CPU= "1 $ncpu 1"       # [PARALLEL] CPUs for each role
DIP_ROLEs= "k c v"         # [PARALLEL] CPUs roles (k,c,v)
DIP_Threads=  0            # [OPENMP/X] Number of threads for dipoles
X_and_IO_CPU= "1 1 1 $ncpu 1"     # [PARALLEL] CPUs for each role
X_and_IO_ROLEs= "q g k c v"       # [PARALLEL] CPUs roles (q,g,k,c,v)
X_and_IO_nCPU_LinAlg_INV= $ncpu   # [PARALLEL] CPUs for Linear Algebra
X_Threads=  0              # [OPENMP/X] Number of threads for response functions
SE_CPU= " 1 $ncpu 1"       # [PARALLEL] CPUs for each role
SE_ROLEs= "q qp b"         # [PARALLEL] CPUs roles (q,qp,b)
SE_Threads=  0    


echo "Running on $ncpu MPI, $nthreads OpenMP threads"
mpirun -np $ncpu  $bindir/yambo -F $filein -J $jdir -C $cdir

As soon as you are ready to submit the job.

$ sbatch

Yambo calculates the GW-qp corrections running on 1 MPI process with a single thread. As you can see, monitoring the log file produced by yambo, the run takes some time, although we are using minimal parameters.

The status of the jobs can be monitored via:

$ squeue  -u $USER        # to inspect the status of jobs 
                          # (hint: make a unix alias, if you like)
$ scancel  <jobid>        # to delete jobs in the queue

Pure MPI scaling with default parallelization scheme

Meanwhile we can run the code in parallel. Let's use consider the case of a node having 16 cores (you can try to adapt the following discussion to the actual maximum number of cores/node you have in your system). As a first run, we'll use 16 MPI tasks, still with a single thread. To this end modify the script changing

#SBATCH --tasks-per-node=16
#SBATCH --cpus-per-task=1

This time the code should be much faster. Once the run is over try to run the simulation also on 2, 4, 8 MPI tasks. Each time, please remember to change both the number of tasks per node both in the header and in the ntasks_per_node variable. Finally, you can try to produce a scaling plot.

To analyze the data you can use the phyton script run which is provided.

You can use it running

$ ./ run*/r-*

You should obtain something like that (but with more columns)

# ncores       dip          Xo           X         io_X       io_WF       Sgm_x        Sgm_c     (REDUX)   WALL_TIME
      1    4.7337s   13m39.00s     0.1500s      0.0241s     0.2487s    34.2143s     15m7.00s     0.0000s      29m29s
      4    1.6019s   218.7982s     0.0882s      0.0283s     0.2077s     9.3338s    242.4438s     0.0001s      07m54s
      8    1.0755s   127.3209s     0.0974s      0.0291s     0.2134s     5.4490s    140.7788s     0.6926s      04m38s
     12    0.7510s    89.1649s     0.1015s      0.0299s     0.2068s     4.2961s    109.1227s     0.0007s      03m26s
     16    0.7653s    68.2550s     0.1048s      0.0309s     0.2463s     2.9211s     72.6220s     0.2799s      02m27s

Plot the execution time vs the number of MPI tasks and check (do a log plot) how far you are from the ideal linear scaling. Below a similar plot produced on a local cluster equipped with two Intel(R) Xeon(R) Silver 4208 CPU @ 2.10GHz processors per node (16 physical cares/node).

Scaling MPI corvina.jpg


  • not all runlevels scale in in the same way
  • you should never overload the available number of cores

Pure OpenMP scaling

Next step is instead to check the OpenMP scaling. Set back

#SBATCH --tasks-per-node=1
#SBATCH --cpus-per-task=16

and now use


Since we are already using 16 threads, we cannot also distribute among MPI tasks, i.e. ncpu will result equal to 1. Try setting nthreads to 16, 8, 4 and 2 and again to plot the execution time vs the number of threads using the python script. Again you should be able to produce data similar to the following. In the following we stopped increasing the number of threads up to 8 because of the specific architecture used in the runs (a dual socket machine with 8 cores/socket).

# ncores   threads         dip          Xo           X         io_X       io_WF       Sgm_x         Sgm_c   WALL_TIME
      1          1     4.7337s   13m39.00s     0.1500s      0.0241s     0.2487s    34.2143s      15m7.00s      29m29s
      2          2     3.1971s   549.1491s     0.2248s      0.0298s     0.2491s    17.6552s     584.3692s      19m17s
      4          4     2.9419s   358.5202s     0.1928s      0.0289s     0.2590s     9.2010s     421.2219s      13m15s
      8          8     2.7992s   344.3342s     0.2332s      0.0325s     0.2543s     5.4334s     362.6982s      11m58s


  • OpenMP usually shares the memory among threads, but not always
  • you should never overload the available number of cores
  • in principle, we could overload the cores setting more threads than the available total number of cores since a single core allows multi-thread operations

MPI vs OpenMP scaling

Which is scaling better ? MPI or OpenMP? How is the memory distributed?

Now you can try running simulations with hybrid strategies. Try for example setting:

#SBATCH --tasks-per-node=8
#SBATCH --cpus-per-task=2
ntasks_per_node= 8
nthreads= 2

We can try to do scaling keeping the total number of threads per node (ntasks_per_node * nthreads) equal to 16. Parsing the data we will obtain something similar to

# ncores         MPI     threads         dip          Xo          X       io_X       io_WF       Sgm_x        Sgm_c  WALL_TIME
      16           2           8     3.6816s   386.0552s    0.0856s    0.0268s     0.2163s    18.9077s    530.0875s     15m43s
      16           4           4     0.9950s    82.5070s    0.1098s    0.0299s     0.2051s     3.0837s    122.6565s     03m32s
      16           8           2     0.8524s    72.6708s    0.0986s    0.0293s     0.2282s     3.0379s     88.1589s     02m48s
      16          16           1     0.7653s    68.2550s    0.1048s    0.0309s     0.2463s     2.9211s     72.6220s     02m27s

As you can see here the total CPU time decreases more and more moving the parallelization from the OpenMP to the MPI level. Sigma_c in particular scales better. Nevertheless, note that the relative performance of the different parallel configurations may strongly depend on the actual machine you are running on.

However, CPU time is not the only parameter you need to check. The total memory usage is also very critical since the GW method may have a large memory footprint. If you have compiled yambo with the flag --enable-memory-profile, the memory usage is tracked and the maximum allocated mem is printed in the report file, and can be extracted typing:

$ grep "Max memory used"  <report_file>

In general yambo can distribute memory when using MPI parallelism (though the actual amount depends on the distribution of MPI tasks across MPI levels). Nevertheless, some memory replication is still present. In general, within the node OpenMP helps in easing the memory usage. Therefore, in cases where the node memori is tight, one may consider changing some MPI tasks for OpenMP threads within the node.

Using a hybrid scheme you may also consider running yambo on mode than one node. To run on two nodes for example you need to set



accordingly you can now set

nthreads= 4

This time you will use 32 cores with (16 per node) 4 OpenMP threads and 2*16/4=8 MPI tasks.


  • in real life calculations running on n_cores > 100, it is a good idea to adopt a hybrid approach
  • with OpenMP, you cannot exit the single node, with MPI you can

Advanced: Comparing different parallelization schemes (optional)

Up to now, we used the default parallelization scheme. Yambo also allows you to tune the parameters which controls the parallelization scheme. To this end, you can open again the script and modify the section where the yambo input variables are set

X_and_IO_CPU= "1 1 1 $ncpu 1"      # [PARALLEL] CPUs for each role
X_and_IO_ROLEs= "q g k c v"        # [PARALLEL] CPUs roles (q,g,k,c,v)
#X_and_IO_nCPU_LinAlg_INV= $ncpu   # [PARALLEL] CPUs for Linear Algebra
X_Threads=  0               # [OPENMP/X] Number of threads for response functions
DIP_Threads=  0             # [OPENMP/X] Number of threads for dipoles
SE_CPU= "1 1 $ncpu"         # [PARALLEL] CPUs for each role
SE_ROLEs= "q qp b"          # [PARALLEL] CPUs roles (q,qp,b)
SE_Threads=  0    

In particular, "X_and_IO_CPU" sets how the MPI Tasks are distributed in the calculation of the response function. The possibilities are shown in the "X_and_IO_ROLEs". The same holds for "SE_CPU" and "SE_ROLEs" which control how MPI Tasks are distributed in the calculation of the response function.

Please try different parallelization schemes and check the performances of Yambo. In doing so, you should also change the jobname in the script


Using the python script, you can then check how speed, memory and load balance between the CPUs are affected. For more details, see also the Parallel module


  • the product of the numbers entering each variable (i.e. X_and_IO_CPU and SE_CPU) times the number of threads should always match the total number of cores (unless you want to overload the cores taking advantage of multi-threads)
  • using the X_Threads and SE_Threads variables, you can think about setting different Hybrid schemes in between the screening and the self-energy runlevel.
  • memory better scales if you parallelize on bands (c v b)
  • parallelization on k-points performs similarly to parallelization on bands, but memory requires more memory
  • parallelization on q-points requires much less communication in between the MPI tasks. It may be useful if you run on more than one node and the internode connection is slow

Prev: Tutorials Home Now: Tutorials Home --> GW Parallel Next: GW Convergence