...
Production environment
Since LEONARDO is a general purpose system and is used by several users at the same time, long production jobs must be submitted using a queuing system (scheduler). The scheduler guarantees that the access to the resources is as fair as possible
The production environment on LEONARDO, is based on the slurm scheduler, already in place on the cluster but still not complete and in a pre-production configuration.
Leonardo is based on a policy of node sharing among different jobs, i.e. a job can ask for resources and these can also be a part of a node, for example few cores and 1GPU. This means that, at a given time, one physical node can be allocated to multiple jobs of different users. Nevertheless, exclusivity at the level of the single core is guaranteed by low-level mechanisms.
Roughly speaking, there are two different modes to use an HPC system: Interactive and Batch. For a general discussion see the section Production Environment.
Interactive
A serial program can be executed in the standard UNIX way:
> ./programThis is allowed only for very short runs on the login nodes. Soon we will impose 10 minutes cpu-time limit for the interactive processes. Please do not execute parallel applications on the login nodes!
Batch
As usual on HPC systems, the large production runs are executed in batch mode. This means that the user writes a list of commands into a file (for example script.x) and then submits it to a scheduler (SLURM for Leonardo) that will search for the required resources in the system. As soon as the resources are available script.x is executed and the results and sent back to the user.
This is an example of script file:
...
- Please refer to the general online guide to slurm and on task/thread bindings, and please pay attention to the setting of the SRUN_CPUS_PER_TASK for hybrid applications dispatched with "srun".
You can write your script file (for example script.x) using any editor, then you submit it using the command:
> sbatch script.xThe script file must contain both directives to SLURM and commands to be executed, as better described in the section Batch Scheduler SLURM.
Using SLURM directives you indicate the account_number (-A: which project pays for this work), where to run the job (-p: partition), what is the maximum duration of the run (--time: time limit). Moreover you indicate the resources needed, in terms of cores, GPUs (later) and memory.
One of the commands will be probably the launch of a parallel MPI application. In this case the right command is srun, as an alternative to the usual mpirun command. In this way you will get full support for process tracking, accounting, task affinity, suspend/resume and other features.
Please note: the "mail" directives are not effective yet.
SLURM partitions
A list of partitions defined on the cluster, with access rights and resources definition, can be displayed with the command sinfo:
> sinfo -o "%10D %20F %P"
The command returns a more readable output which shows, for each partition, the total number of nodes and the number of nodes by state in the format "Allocated/Idle/Other/Total".
In the following table you can find the main features and limits imposed on the partitions of Leonardo.
...
SLURM
partition
...
max running jobs per user/
max n. of nodes/cores/GPUs per user
...
lrd_all_serial
(default)
...
max = 4 physical cores
(8 logical cpus)
max mem = 30800 MB
...
min = 65 nodes
max =256 nodes
...
max = 3 nodes
...
- For EUROFusion users and their dedicated queues please refer to the dedicated document.
Graphic session
It will be available soon.
Programming environment
Leonardo compute nodes host four A100 GPUs per node (CUDA compute capability 8.0). The most recent versions of nVIDIA CUDA toolkit and of the nVIDIA nvhpc compilers (ex PGI, supporting CUDA Fortran) is available in the module environment.
Compilers
You can check the complete list of available compilers on Leonardo with the command:
> module availableand checking the "compilers" section. The available compilers are:
- Gnu Compilers Collection (GCC)
- NVIDIA nvhpc (ex PGI)
- CUDA
NVIDIA nvhpc (ex PORTLAND PGI + NVIDIA CUDA)
As of August 5, 2020, the "PGI Compilers and Tools" technology is a part of the NVIDIA HPC SDK product, available as a free download from NVIDIA.
...
For legacy reasons, the nVIDIA nvhpc suite also offers the PGI C, C++, and Fortran compilers with their original names:
...
To enable CUDA C++ or CUDA Fortran, and link with the CUDA runtime libraries, use the -cuda option (-Mcuda is deprecated). Use the -gpu option to tailor the compilation of target accelerator regions.
The OpenACC parallelization is enabled by the -acc flag. GPU targeting and code generation can be controlled by adding the -gpu flag to the compiler command line.
The OpenMP parallelization is enabled by the -mp compiler option. The GPU offload via OpenMP is enabled by the -mp=gpu option.
GNU compiler collection
The gnu compilers are always available. GCC version 8.5.0 is available without the need to load any gcc module. In the module environment you can find more recent version though.
The name of the GNU compilers are:
- g77: Fortran77 compiler
- gfortran: Fortran95 compiler
- gcc: C compiler
- g++: C++ compiler
The documentation can be obtained with the man command after loading the gnu module:
> man gfortan
> man gcc
CUDA
Compute Unified Device Architecture is a parallel computing platform and programming model developed by NVIDIA for general computing on graphical processing units (GPUs). With CUDA, developers are able to dramatically speed up computing applications by harnessing the power of GPUs.
In GPU-accelerated applications, the sequential part of the workload runs on the CPU – which is optimized for single-threaded performance – while the compute intensive portion of the application runs on thousands of GPU cores in parallel. When using CUDA, developers program in popular languages such as C, C++, Fortran, Python and MATLAB and express parallelism through extensions in the form of a few basic keywords. We refer to the NVIDIA CUDA Parallel Computing Platform documentation.
Debugger and Profilers
If at runtime your code dies, then there is a problem. In order to solve it, you can decide to analyze the core file (core not available with PGI compilers) or to run your code using the debugger.
Compiler flags
Whatever your decision, in any case, you need to enable compiler runtime checks, by putting specific flags during the compilation phase. In the following we describe those flags for the different Fortran compilers: if you are using the C or C++ compiler, please check before because the flags may differ.
The following flags are generally available for all compilers and are mandatory for an easier debugging session:
-O0 Lower level of optimization -g Produce debugging information
Other flags are compiler specific and are described in the following:
PORTLAND Group (PGI) Compilers
The following flags are useful (in addition to "-O0 -g") for debugging your code:
-C Add array bounds checking
-Ktrap=ovf,divz,inv Controls the behavior of the processor when exceptions occur:
FP overflow, divide by zero, invalid operands
GNU Fortran compilers
The following flags are useful (in addition to "-O0 -g")for debugging your code:
-Wall Enables warnings pertaining to usage that should be avoided
-fbounds-check Checks for array subscripts.Debuggers available
GNU: gdb (serial debugger)
GDB is the GNU Project debugger and allows you to see what is going on 'inside' your program while it executes -- or what the program was doing at the moment it crashed.
VALGRIND
Valgrind is a framework for building dynamic analysis tools. There are Valgrind tools that can automatically detect many memory management and threading bugs, and profile your programs in detail. The Valgrind distribution currently includes six production-quality tools: a memory error detector, two thread error detectors, a cache and branch-prediction profiler, a call-graph generating cache profiler, and a heap profiler.
Valgrind is Open Source / Free Software, and is freely available under the GNU General Public License, version 2.
Profilers
In software engineering, profiling is the investigation of a program's behavior using information gathered as the program executes. The usual purpose of this analysis is to determine which sections of a program to optimize - to increase its overall speed, decrease its memory requirement or sometimes both.
A (code) profiler is a performance analysis tool that, most commonly, measures only the frequency and duration of function calls, but there are other specific types of profilers (e.g. memory profilers) in addition to more comprehensive profilers, capable of gathering extensive performance data.
gprof
The GNU profiler gprof is a useful tool for measuring the performance of a program. It records the number of calls to each function and the amount of time spent there, on a per-function basis. Functions which consume a large fraction of the run-time can be identified easily from the output of gprof. Efforts to speed up a program should concentrate first on those functions which dominate the total run-time.
gprof uses data collected by the -pg compiler flag to construct a text display of the functions within your application (call tree and CPU time spent in every subroutine). It also provides quick access to the profiled data, which let you identify the functions that are the most CPU-intensive. The text display also lets you manipulate the display in order to focus on the application's critical areas.
Usage:
> gfortran -pg -O3 -o myexec myprog.f90
> ./myexec
> ls -ltr
.......
-rw-r--r-- 1 aer0 cineca-staff 506 Apr 6 15:33 gmon.out
> gprof myexec gmon.out
It is also possible to profile at code line-level (see "man gprof" for other options). In this case, you must use also the “-g” flag at compilation time:
> gfortran -pg -g -O3 -o myexec myprog.f90
> ./myexec
> ls -ltr
.......
-rw-r--r-- 1 aer0 cineca-staff 506 Apr 6 15:33 gmon.out
> gprof -annotated-source myexec gmon.out
It is possible to profile MPI programs. In this case, the environment variable GMON_OUT_PREFIX must be defined in order to allow to each task to write a different statistical file. Setting
export GMON_OUT_PREFIX=<name>once the run is finished each task will create a file with its process ID (PID) extension
<name>.$PIDIf the environmental variable is not set every task will write the same gmon.out file.
Nvidia Nsight System (GPU profiler)
Nvidia Nsight System is a system-wide performance analysis tool designed to visualize an application’s algorithms, help you identify the largest opportunities to optimize, and tune to scale efficiently across any quantity or size of CPUs and GPUs; from large server to our smallest SoC.
You can find general info on how to use it in the dedicated Nvidia User Guide pages.
Our suggestion is to run the CLI inside your job script in order to generate the qdrep files. Then you can download the qdrep files on your local PC and visualize them with the Nsight System GUI available on your workstation.
The profiler is available under the module nvhpc.
Standard usage of an MPI job running on GPU is
> mpirun <options> nsys profile -o ${PWD}/output_%q{OMPI_COMM_WORLD_RANK} -f true --stats=true --cuda-memory-usage=true <your_code> <input> <output>On the single node you can also run the profiler as "nsys profile mpirun", but keep in mind that with this syntax nsys will put everything in a single report.
Unfortunately nsys usually generates several files in /tmp dir of the compute node even if a TMPDIR environment variable is set. These files may be big causing the filling of the /tmp folder and, as a consequence, the crash of the compute node and the failure of the job.
In order to avoid such a problem we strongly suggest to include in your sbatch script the following lines around your mpirun call as a workaround:
> rm -rf /tmp/nvidia
> ln -s $TMPDIR /tmp/nvidia
> mpirun ... nsys profile ...
> rm -rf /tmp/nvidia
This will place the temporary outputs of the nsys code in your TMPDIR folder that by default is /dev/shm/slurm_job.$SLURM_JOB_ID where you have about 250 GB of free space.
This workaround may cause conflicts between multiple jobs running this profiler on a compute node at the same time, so we strongly suggest also to request the compute node exclusively:
#SBATCH --exclusiveMPI environment
We offer two options for MPI environment on LEONARDO:
- Open MPI
- Intel-OneAPI-MPI
Here you can find some useful details on how to use them on LEONARDO.
Compiling
OpenMPI
This most common MPI implementation is installed inside the GNU environment.
It is configured to support CUDA-aware.
To install MPI applications using Open MPI you have to load openmpi module (use "modmap -m openmpi" command to see the available Open MPI versions) and select the MPI compiler wrapper for Fortran, C or C++ codes.
The openmpi module provides the following wrappers:
...
Compiler
...
Wrapper
...
Usage
...
mpic++
mpiCC
mpicxx
...
mpif77
mpif90
mpifort
...
e.g. Compiling C code
> module load openmpi/<version>
> mpicc -o myexec myprog.c (uses the gcc compiler)
You can add all options available for the backend compiler (you can show it by "-show" flag, e.g. "mpicc -show"). In order to list them type the "man" command:
> man mpiccIntel-OneAPI-MPI
This is the MPI implementation of Intel and doesn't support CUDA.
To install MPI applications using Intel MPI you have to load intel-oneapi-mpi module (use "modmap -m intel-oneapi-mpi command to see the available versions).
The intel-oneapi-mpi module provides the following wrappers for classic intel compilers and oneapi ("x") compilers:
...
Compiler
...
Wrapper
...
Usage
icpc
icpx
mpiicpc
mpiicpc -cxx=icpx
...
Compile C++ source files with classic Intel
Compile C++ source files with oneapi
icc
icx
mpiicc
mpiicc -cc=iccx
...
Compile C source files with classic Intel
Compile C source files with oneapi
ifort
ifx
mpiifort (Fortran90/77)
mpiifort -fc=ifx
...
Compile FORTRAN source files with classic Intel
Compile FORTRAN source files with oneapi
e.g. Compiling Fortran code
> module load intel-oneapi-mpi/<version>
> mpiifort -o myexec myprog.f90 (uses the ifort compiler)
You can add all options available for the backend compiler (you can show it by "-show" flag, e.g. "mpicc -show"). In order to list them type the "man" command
> man mpiifortRunning
To run MPI applications they are two way:
- using mpirun launcher
- using srun launcher
mpirun launcher
To use mpirun launcher the openmpi or intel-oneapi-mpi module needs to be loaded:
> module load openmpi/<version>or
> module load intel-onepi-mpi/version...
It can be used via salloc or sbatch way:
> salloc -N 2 (allocate a job of 2 nodes)
> mpirun ./mpi_exec
or
> sbatch -N 2 my_batch_script.sh (allocate a job of 2 nodes)
> cat my_batch_script.sh
#!/bin/sh
mpirun ./mpi_exec
srun launcher
MPI applications can be launched directly with the slurm launcher srun
> srun -N 2 ./mpi_execor via salloc/sbatch way:
> salloc -N 2 (allocate a job of 2 nodes)
> srun ./mpi_exec
or
> sbatch -N 2 my_batch_script.sh (allocate a job of 2 nodes)
> vi my_batch_script.sh
#!/bin/sh
srun -N 2 ./mpi_exec
Scientific libraries
Linear Algebra
GPU accelerated
The nvidia math libraries are available by loadind "nvhpc" module (use "modmap -m nvhpc" command to see the available versions of nvhpc).
For not nvidia math libraries installed with cuda support they are available by loading the corresponding module e.g "module load magma/<vers>". Notice that when you load the module of any of these libraries the CUDA module is not automatically loaded).
- BLAS: nvidia cublas, magma
- LAPACK: nvidia cusolver, magma
- SCALAPACK: slate
- EIGENVALUE SOLVERS: nvidia cusolver, magma (single-node), slate, elpa and slepC (multi-node)
- SPARCE MATRICES : nvidia cuSPARSE, PetSc (multi-node), SuperLU-dist (multi-node)
- Hypre (multi-node)
CUDA not supported
- BLAS: openblas, intel-oneapi-mkl
- LAPACK: openblas, intel-oneapi-mkl
- SCALAPACK: netlib-scalapack, intel-oneapi-mkl
Fast Fourier Transform
GPU accelerated
The nvidia math libraries are available by loadind "nvhpc" module (use "modmap -m nvhpc" command to see the available versions of nvhpc).
nvidia cuFFT/cuFFTW (single-node)
CUDA not supported
- FFTW (single and multi-node)