SIESTA Deployment Options¶
- Author
Vladimir Dikan <vdikan@icmab.es> (ICMAB-CSIC)
Table of Contents
As you might know (or will realize during the school), SIESTA has quite a number of capabilities and operation modes. And and as many other HPC codes, it relies on quite a number of dependencies and build options, that sometimes give their users hard times with configuration of their research environments.
Below is a review of some aspects of compilation and deployment of SIESTA. Some pre-configured options are discussed, followed by an overview of SIESTA’s general Makefile template and dependencies for a manual compilation. Finally, a couple of scripted provisioning systems are mentioned to give users an idea of how building if SIESTA can be streamlined for different environments.
Before continuing please, remember to first and foremost address the SIESTA manual and in particular the “Compilation” section for instructions on installation of SIESTA.
Also please keep in mind that this is not an exhaustive overview of all possible installation options for SIESTA. After the recent migration of its codebase to the GitLab repository, other installation and packaging systems are developed further to provide deployment options for SIESTA. Stay updated with the SIESTA project announcements!
1 Ready-to-Use Options¶
For personal use, either just to try siesta or to run standard and not very complex jobs, there are virtualized “plug-and-play” variants available:
QuantumMobile
virtual machine with a number of computational codes pre-installed, along with pre-configured for themAiiDA
instance.Conda installation: See this how-to.
Containers for different platforms such as
Docker
andSingularity
1.1 QuantumMobile VM¶
The QuantumMobile virtual machine is set up with pre-installed SIESTA, as well as with a number of other computational codes curated by MARVEL and MaX European initiatives. It may be the easiest installation option for beginners. In fact, QuantumMobile will be used during the SIESTA school for its students’ accounts.
Note
It’s recommended to use VirtualBox 6.1 with QuantumMobile.
[Screenshots not included in this version]
Download the image from QuantumMobile Releases page, launch VirtualBox
and select the .ova
file in the Import Appliance menu.
After launching the VM one can see that siesta
is available in the
terminal emulator, along with some of its useful utilities like
tbtrans
, vibra
, gnubands
, denchar
etc.
Moreover, after launching AiiDA activation command (as the splash
screen suggests) we obtain AiiDA’s verdi
shell with siesta
code already pre-configured to be used with AiiDA.
The header of SIESTA output shows the code configuration options and dependencies with which the code was built inside the VM.
1.2 Containers¶
When a more lightweight variant of virtual environment is preferred, the containers such as Docker and Singularity can be used. We provide experimental containers for SIESTA, and plan to auto-generate them for future releases.
1.2.1 Docker¶
Docker images for SIESTA can be found on this Docker Hub page. Sample usage:
docker pull vdikan/siesta-dist:master
docker run --interactive --tty -w /app -v "$(pwd):/app" vdikan/siesta-dist:master
Assuming one has all required input in the current local directory,
the command above will launch a container for siesta@master
development branch, mount said directory as a volume under /app
path
and pass control of the terminal inside the container to the user.
Executables from the Siesta package will be visible and should work
while inside the container, and after shutdown the local directory
mounted as /app
should persist together with all the generated output.
1.2.2 Singularity¶
The Singularity project is an alternative to Docker aimed primarily for scientists and their supercomputers. At a glance, using Singularity is more convenient because it helps running containerized software in a manner more familiar to researchers (that is, running binaries within a scope of some directory path).
It is arguably most convenient to convert existing Docker images to Singularity:
singularity build siesta-dist.sif docker://vdikan/siesta-dist:master
Will produce a Singularity container image from the Docker image from the previous section, and
singularity run siesta-dist.sif
will launch it in a way similar to the Docker container.
Note
There is no need to specifically designate and mount a volume with the current folder so that the results of calculations are saved: Singularity does so automatically.
2 Source Code Compilation¶
The detailed instructions how to build SIESTA from source code are written in the SIESTA manual. Here some details are clarified.
2.1 Staging the build directory¶
It’s recommended to obtain stable versions of SIESTA from the Siesta Project releases page, e.g.:
wget https://gitlab.com/siesta-project/siesta/-/archive/v4.1.5/siesta-v4.1.5.tar.gz
tar -xzf siesta-v4.1.5.tar.gz
cd siesta-v4.1.5/
ls
AUTHORS COPYING Docs Examples NOTICE.txt Obj Pseudo README.md ReleaseNotes.md Src Tests Tutorials Util version.info
The source code is placed in Src/
directory, but as the manual
states, compilation inside Src/
is not allowed and will cause an
error:
cd Src/
make
** You can no longer build SIESTA in Src.
** Go to the Obj directory and see the README file.
Makefile:62: recipe for target 'what' failed
make: *** [what] Error 1
The correct way to build SIESTA is to use a separate build directory.
The default such directory is present in the archive and is called
Obj/
, but users may have any number of build directories on the same
level, with different configurations of SIESTA.
First, the build directory should be staged by calling a setup script:
cd Obj/
sh ../Src/obj_setup.sh
*** Compilation setup done.
*** Remember to copy an arch.make file into the directory.
*** These files are template arch.make files:
*** gfortran.make (for gfortran compiler)
*** intel.make (for intel compiler)
*** DOCUMENTED-TEMPLATE.make (requires customization)
Then an arch.make
file should be placed with options that customize
the build. The output of previous command lists a few templates
located in Obj/
. New releases also come with a number of well-structured
and self-documented arch.make
templates located in
Obj/ARCH_EXPERIMENTAL/
tree Obj/
Obj/
├── ARCH-EXPERIMENTAL
│ ├── cte-gcc.mk
│ ├── cte-ibm.mk
│ ├── gcc-modules.mk
│ ├── master-raw.make
│ ├── mn-intel.mk
│ └── README
├── DOCUMENTED-TEMPLATE.make
├── gfortran.make
├── intel.make
└── README
It is recommended to use these new templates to compile SIESTA.
cp ARCH-EXPERIMENTAL/master-raw.make ./arch.make
$EDITOR arch.make
2.2 arch.make
configuration¶
The structure of the new arch.make
templates is very
self-explanatory. Users should edit only the header part before this
line:
#--------------------------------------------------------
# Nothing should need to be changed below
#--------------------------------------------------------
First section specifies internal flags for SIESTA dependencies. Set
them to 1
for the required options, e.g. WITH_MPI=1
The rest leave blank.
The next section defines linking symbols for mandatory dependencies as well for some external libraries marked as required. When requesting any of them, uncomment the corresponding line and put the appropriate library symbols. Please, contact your system administrator to find out the correct library locations on a shared system.
Example: netcdf
dependency switched on with separate
netcdf-fortran
interface:
WITH_NETCDF=1
WITH_SEPARATE_NETCDF_FORTRAN=1
WITH_NCDF=1
...
NETCDF_ROOT=$(NETCDF_HOME) # /path/to/netcdf-c
NETCDF_FORTRAN_ROOT=$(NETCDF_HOME) # /path/to/netcdf-fortran
Finally, specify compiler options that aren’t visible or do differ from the shell environment (alternative ways possible, see the inline documentation commentaries).
FC_PARALLEL=mpif90
FC_SERIAL=gfortran
FPP = $(FC_SERIAL) -E -P -x c
FFLAGS = -O2
FFLAGS_DEBUG= -g -O0
RANLIB=echo
...
2.3 Siesta Dependencies¶
The dependencies for SIESTA can be divided in two groups: common ones that are widely used for many projects in HPC and computational research, and specific for SIESTA project: libraries used in siesta program (almost) exclusively, written by authors of SIESTA. The majority of SIESTA-specific dependencies are grouped on the corresponding section on the project page: https://gitlab.com/siesta-project/libraries
Note
SIESTA code already ships with minimum necessary
libraries. In principle, even use of MPI
is optional: SIESTA can be
built and run in serial mpde. The only requirement is installed
SCALAPACK
library in case when MPI
is requested.
The following table lists some external libraries that can be linked with SIESTA and extend its capabilities:
Common |
SIESTA-specific |
---|---|
MPI, BLAS, LAPACK |
|
ELSI + ext. solvers |
|
2.4 Building Utilities¶
Many useful features of SIESTA are extracted as separate utilities in
the Util/
library. One can build them separately or perform a batch
compilation with Util/build_all.sh
.
Note
Utilities rely on the arch.make
configuration for SIESTA.
Build them after successful compilation of the siesta
executable in
Obj/
, and/or edit the corresponding Makefile
-s accordingly
(paying attention to environment variables pointing to siesta
, e.g.
OBJDIR
). Then proceed with:
cd Util && ./build_all.sh
3 Scripted Installations¶
Accurate management of dependencies and build environment for SIESTA (or any comparably complex scientific code) can be cumbersome. That is why several scripted systems exist that can aid setting up of said environment.
Those systems and packages differ in complexity and operation quality,
being either a set of Make
-scripts, Git
actions, or a set of
packages for software managers such as Spack
. Users and
system administrators have a choice of options in cases where a
1 Ready-to-Use Options solution does not suite.
A couple of variants are described in this section.
3.1 Siesta-Install-Scripts¶
A lightweight set of scripts for manual installation of siesta and its dependencies. Authored by Alberto Garcia and Xe Hu.
The project repository contains branches adapted for different operating
systems and HPC clusters; e.g. users of Debian/Ubuntu
can use
this branch. For documentation address the README
files and inline
documentation comments.
git clone https://gitlab.com/mailhexu/siesta-install-scripts.git
cd siesta-install-scripts/
# 1. address README-s in subdirectories
# 2. inspect ./Tarballs/download.sh
./Tarballs/download.sh
cp cp Config/gnu/* Tarballs/ # copy Makefile for GNU target platform
cp Config/siesta.common.arch.make Tarballs/
# 3. inspect and adapt Makefiles
# 4. inspect and edit ./do_all.sh
./do_all.sh
3.2 Spack Siesta package¶
It is possible to install SIESTA with experimental Spack package. Spack is a package manager targeting research software and supercomputers, although it proves useful even for software management on a personal machine. Please consult Spack documentation for installation instructions.
Warning
In order to get Spack siesta package for now one needs to clone
this dedicated branch of spack
(called siesta-develop
), not the core spack repository! The rest of
installation described in the docs is valid for said branch as for
core spack v.15.4-16.0
We are working towards providing a single one-liner command to install SIESTA releases with a single one-line specification for Spack. At the moment you can build experimental versions of SIESTA with Spack obtained through:
git clone -b siesta-develop https://github.com/vdikan/spack.git
After configuration of compilers for Spack, the installation of SIESTA is done in principle with a single spec command, e.g.:
spack install siesta@master +utils ^openmpi +cxx +cxx_exceptions
installs parallel version of siesta from GitLab’s @master branch with
mpi provided by openmpi, with C++ support, as well as key siesta
utilities like denchar
, tbtrans
and stm
.
In order to make the executables available either run:
spack load -r siesta@master # "@" precedes the version installed by spack
or address Spack’s built-in modulefiles generation mechanism.
3.2.1 Available Spack versions of SIESTA:¶
At the moment there are few SIESTA versions visible for the experimental Spack package, namely:
@master - for the master branch of the project on GitLab
@psml - for the branch with PSML pseudopotentials support (downloaded from Git)
@elsi - for the branch with ELSI+PEXSI support (downloaded from Git). Requires MPI built with cxx, as in the example above.
@4.1-b4 -for the stable version hosted on Launchpad