23
$\begingroup$

In some science fields there are efforts to provide customized distributions, so you can get a productive environment with minimum set-up overhead. For example, in high energy particle physics, we have Scientific Linux, in mathematics we have MathLibre, and in cybersecurity research we have Kali Linux.

Is there any effort like these for materials modeling? If not, which Linux distributions come the closest?

Please list only one Linux distribution per answer!

$\endgroup$
  • 4
    $\begingroup$ What would the difference be form Scientific Linux? Materials Modelling and Particle-Physics Modeling are essentially the same in terms of the types of calculations being done: lots of heavy numerical computations making use of libraries like BLAS/MKL and lots of parallel HPC computing or GPU computing. $\endgroup$ – Nike Dattani Jun 8 at 18:22
  • 1
    $\begingroup$ @CodyAldaz, I was thinking more about open source/free alternatives, but for reference porposes I think these would be valid answers too. NikeDattani, I still didn't test Scientific Linux as a base system, but I'm considering doing just so, and opened this question before to make sure I'm not missing any other alternatives that may exist out there. $\endgroup$ – ksousa Jun 8 at 18:54
  • 2
    $\begingroup$ @ksousa that's fair. Well this might be an XY question where you ask for X but what you really need is Y. I guess you want to know: as someone planning to do a lot of matter modeling calculations, which Linux distribution is the best? The answer might be Scientific Linux, it might be something even more specific to Matter Modeling, or it might even just be Fedora. If you ask it that way (which Linux distribution is best for the following Matter Modeling things I want to do), I'm sure we'll get 5 answers giving the advantages and disadvantages of each, but the way it is, maybe 0 answers. $\endgroup$ – Nike Dattani Jun 8 at 19:37
  • 1
    $\begingroup$ By the way, I was NOT the downvoted. I actually upvoted. You can see that I do not yet have the "critic badge" meaning that I have never downvoted yet, on this site. $\endgroup$ – Nike Dattani Jun 8 at 19:38
  • 2
    $\begingroup$ This will probably become a Hot Network Question, so I've shortened it to one line. $\endgroup$ – Nike Dattani Jun 8 at 23:12
15
$\begingroup$

Fedora

Susi Lehtola's answer makes some good points, and I will elaborate on two of them:

  • The Linux distributions mentioned in your question, which cater to specific communities (e.g. Scientific Linux for particle physics, MathLibre for math, Kali Linux for cybersecurity) are descendants of much more widely-known distributions, so they are not acutely different from what a lot of people are already familiar with using. Susi mentioned that Scientific Linux is a descendant of Red Hat, and I'll add to that by saying that both MathLibre and Kali are direct descendants of KNOPPIX, which descended from Debian between 2001-2002, a few years before Ubuntu descended from Debian. So of the three distributions you mentioned: 2 are very closely related to Debian/Ubuntu, and 1 is very closely related to RedHat/CentOS/Fedora. This would likely also be the case with anything else catering specifically to matter modeling.
  • "If the distribution has a contributor who needs package X for their interests, it's highly likely that the distribution's package X is up to date and includes all features." In light of this, since Susi has been a packager for Fedora for over a decade and packaged OpenMOLCAS, PyQuante, Psi4, PySCF, linint, libxc, and whatever else (he can edit this if there's more!) for Fedora, then Matter Modelers might find Fedora/RedHat/CentOS to be convenient since so many of our workhorses have already been packaged for Fedora.

To focus this answer on just one Linux distribution, out of the Fedora/RedHat/CentOS family, I picked Fedora because:

  • It's free. (Red Hat Enterprise Linux comes with official support, but costs money).
  • It caters to desktop users. (CentOS is also free, but it caters more to servers, so it misses a lot of the packages for desktop use, so the Fedora user community is more likely to grow to a much larger size; Fedora also caters a bit to servers and cloud computing, so it's versatile in that way).
  • It's updated frequently. (CentOS and RHEL are preferred for servers because they are very stable and well-tested, but this is accomplished by not updating the all the latest packages).
  • The people I've met that use Fedora, are extremely enthusiastic about it (so the even if the number of users is not the biggest, the users love Fedora, will continue to promote and develop Fedora, and won't leave Fedora any time soon), and also very welcoming for others to joint the community.
  • There's about 1.5 million users, including the inventor of Linux: Linus Torvalds.

Drawbacks:

  • I mentioned Fedora because your comments suggested to me that you were looking for something for your personal computer. Most matter modeling calculations are done on servers though, and that's when CentOS is used more commonly, largely because it's more stable and tested than Fedora, and the latest fancy packages for which you would want Fedora on your desktop, are not usually needed on a server.
  • As a desktop OS, Fedora still has a much smaller user community than Ubuntu which has an estimated 20 million users. Having too large a user community also leads to some disadvantages to Ubuntu, but they have an entire StackExchange site of their own, so when troubleshooting you might get an answer faster, and also I personally have found it helpful when giving talks at other universities (using my own laptop) that people at the host university that needed to help me with something, or wanted to set me up to use their printer, were familiar with my Ubuntu OS.
| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ What kept me away from Fedora/RedHat family in the past is what appears to be a short blanket situation, where either you get feet or head covered, but not both. Either you have Fedora short 6-month cycles, get latest software and drivers, but face occasional stability issues, or you have RedHat/CentOS 5-year cycles, get top stability, but face outdated packages and recent hardware not working out-of-the-box. My impression is that Ubuntu and Debian (included firmware version) got this compromise better. $\endgroup$ – ksousa Jun 11 at 16:21
  • $\begingroup$ Anyway, I believe my opinion is based in outdated information, since the last time I gave a serious try at Fedora family was around 5 years ago. Once I'm used to something, I tend to stick to it. Distro hopping can be a time sink. But once in a while is good to reevaluate your position. $\endgroup$ – ksousa Jun 11 at 16:26
  • 1
    $\begingroup$ Can you confirm for us whether you're asking in regards to a cluster or a personal laptop/desktop? I've never seen a Debian/Ubuntu cluster. The Chemistry Department at Oxford used CentOS, and all of Compute Canada uses CentOS. I'm an Ubuntu user but now that I've found out that Susi Lehtola has packaged many of the Matter Modeling codes that are most important to me, for Fedora/RHEL/CentOS, it does seem like a good answer if you're looking for $\endgroup$ – Nike Dattani Jun 11 at 19:26
  • 1
    $\begingroup$ @ksousa That totally makes sense. You might also have seen that someone asked here about setting up a "cluster" using several connected Raspberry Pi devices. For a personal computer I think the answer is basically Ubuntu or Fedora. Fedora has the advantage that Susi Lehtola has packaged the major open-source electronic structure codes for Fedora, and it seems more stable than when you had that experience a few years ago. Ubuntu has the advantage that more software might be available, and a larger user community (but that's a security issue too) $\endgroup$ – Nike Dattani Jun 11 at 21:06
  • 1
    $\begingroup$ Thanks for accepting mine :) :) I'm a Ubuntu user, so you know my answer was the least possible "biased". $\endgroup$ – Nike Dattani Jun 11 at 21:10
24
$\begingroup$

There isn't a silver bullet for this, so in my humble opinion there isn't a "best" Linux distribution for this. Just as Kali is not "the best" for security and Scientific Linux is also not advertised as the "best" for HEP. They are just good enough alternatives that tackle some issues that may seem helpful for many users, such as including many things and not having to install them by hand.

Personally, I tend to use popular big distributions, that I know are well documented and won't cease to exist anytime soon. Any Linux distribution could be used for Matter Modelling. The features of the software and tools for Modelling are independent of the choice of distribution (in general). Also take into account that a lot of Matter Modelling involves generic tools that are not specific for the area (shell, modern python versions, up-to-date BLAS libraries, etc.).

| cite | improve this answer | |
$\endgroup$
14
$\begingroup$

Ubuntu

For your workstation or laptop the best linux distribution is Ubuntu. Many of the recommendations here, like CentOS, are great for servers, but you might also have a laptop or desktop computer that you use for developing your code and running simulations. Ubuntu has gone much further than other linux distributions in developing a consumer-grade user interface. It's the only linux distribution that comes close to being a user-friendly desktop operating system in the same way as Mac OS or Windows. It's broad user base also means that little things are more likely to work, like your wifi, or your laptop going to sleep when you close the lid.

| cite | improve this answer | |
$\endgroup$
14
$\begingroup$

CAELinux

(to simulate continuous material)

CAELinux's official page states:

Based on the open-source CAD/CAM software such as Freecad, LibreCAD, PyCAM and Slic3r and CAE softwares like Salomé, Code_Aster, Code_Saturne, Calculix, OpenFOAM and Elmer , you can design your CAD geometry, perform multiphysics simulations to optimize your design, generate G-code for prototyping with 3D printing & milling, and even develop your own PCBs & microcontroller based electronic circuits for automation.

| cite | improve this answer | |
$\endgroup$
  • 2
    $\begingroup$ +1. Wow! Another Linux specifically for science/engineering! It seems to be for classical materials simulations (not quantum), and more for engineering than for molecular science, but just like Scientific Linux which is for particle physics, it might have a lot of the things that other materials scientists would be interested in. $\endgroup$ – Nike Dattani Jun 8 at 21:14
  • 2
    $\begingroup$ Yes @NikeDattani, I just added for the multiphysics simulations feature (that's what COMSOL do!) $\endgroup$ – Camps Jun 8 at 21:18
  • $\begingroup$ Good. I will wait a bit before accepting a answer, to see if other suggestions show up. $\endgroup$ – ksousa Jun 8 at 23:12
  • $\begingroup$ Is CAELinux just Xubuntu with more preinstalled software? $\endgroup$ – curiousdannii Jun 9 at 5:08
  • 3
    $\begingroup$ CAElinux looks like a great idea executed poorly, but it's probably worth checking out. Do you have actual experience with it? $\endgroup$ – Mast Jun 9 at 7:13
12
$\begingroup$

Scientific Linux is a rebuild of Red Hat Enterprise Linux. It started out before the CentOS project really took off. Nowadays CentOS is officially supported by Red Hat, but Scientific Linux still exists on its own.

Since both CentOS and Scientific Linux are just rebuilds of RHEL, all three should be binary compatible, and you should be able to use the same software repositories for all three.

The Fedora project is where development of RHEL happens, and the Fedora project also ships software for RHEL releases that haven't been included in the official Red Hat package selection. There is a fair number of matter modeling packages in the Fedora Extra Packages for Enterprise Linux (EPEL) repositories.

However, the problem with RHEL is that since the distribution is "stable", with the current pace of development the system packages quickly become outdated. The widest and most up to date selection is within Fedora, which aims at a new stable release every 6 months.

IMNSHO there are only two distributions that can be considered at earnest for general usage: Debian/Ubuntu, and Fedora/Red Hat (and its clones like Scientific Linux and CentOS), since they are the most widely used and offer by far the widest software selections. (Knoppix is a live CD version of Debian.)

In my personal experience the Debian world tends to lag a bit behind, but it should be noted I am not impartial: I've been packaging for Fedora/Red Hat for over a decade, by addressing any deficiencies wherever I've noticed them; this is why my needs are best served by Fedora and Red Hat.

In any case, the situation dependends highly on the package and its maintainer. If the distribution has a contributor who needs package X for their interests, it's highly likely that the distribution's package X is up to date and includes all features. If not, then it might be that X is not included in the distro, or that the version of X is super old...

| cite | improve this answer | |
$\endgroup$
  • 2
    $\begingroup$ +1 from a package developer to an active package maintainer $\endgroup$ – Geoff Hutchison Jun 9 at 20:24
  • 1
    $\begingroup$ IIRC the origin of Scientific Linux was for a long-term supported stable distro for high-energy physics, correct? The main problem I've encountered with CentOS and SL are old versions of compilers. $\endgroup$ – Geoff Hutchison Jun 9 at 20:24
  • 1
    $\begingroup$ Physicists are very conservative. We write code and want it to be stable for a long time, and care less about having the most up-to-date compilers. This is an advantage in some contexts, but obviously a disadvantage if you want to use the most recent version of a programming language. $\endgroup$ – taciteloquence Jun 10 at 2:48
  • 1
    $\begingroup$ The comment of taciteloquence says that we don't care so much about updating compilers, but I don't quite agree with that: most clusters (especially if used by a large population) do update compilers reasonably often. $\endgroup$ – Nike Dattani Jun 10 at 13:02
  • 1
    $\begingroup$ "there are only two distributions that can be considered at earnest for general usage: Debian/Ubuntu, and Fedora/Red Hat" What about SUSE? The clusters (and all desktops!) I've used at a major Max Planck Institute in Germany, use it. $\endgroup$ – Nike Dattani Jun 10 at 23:05
8
$\begingroup$

VigyaanCD

As a historic note, I found an example of a effort in this direction, in the past. A modified KNOPPIX targeted at molecular modeling with Gamess-US. It was called the VigyaanCD. Unfortunately it appears to be a defunct project by now. It was released in 2005 and I could not find any updates since then. The Gamess site has the following description:

GAMESS, Vigyaan CD (obsolete):

This is a bootable CD, containing a 2005-vintage version of GAMESS. This CD, from the Vigyaan project, is a live Linux CD containing all the required software to boot the computer, and run a standard, but now quite old, version of GAMESS, Avogadro-GMS, GTK-GAMESS and other useful software. Tutorials are also available on this CD. Note, this CD does not require compilation of GAMESS. More information about the CD can be found on the Vigyaan web site. If you encounter any problems with this CD, please do not send email to Mike or our Webmaster, instead send it to Pratul Agarwal.

Perhaps someone will bring it back from the dead, someday.

| cite | improve this answer | |
$\endgroup$
  • 4
    $\begingroup$ +1. Kudos for eventually finding something that was targeted at matter modeling. I see it came with Avogrado-GMS, so I wonder if Geoff knew about it (he didn't mention this software as an answer). $\endgroup$ – Nike Dattani Jun 11 at 15:44
  • 2
    $\begingroup$ Good work ksousa. We have an answer, even though its defunct. $\endgroup$ – Thomas Jun 11 at 15:57
  • $\begingroup$ @Thomas I think basically Fedora can be considered the new VigyaanCD, since so many programs have been packaged by Susi Lehtola. It also has the advantage that it has 1.5 million users, so you can also rely on it having up-to-date non-scientific software too. Basically it covers the best of both worlds. The very related RHEL/CentOS is also optimized for server usage, so Fedora is good in 3 ways. I'm saying this even though I'm an Ubuntu user. $\endgroup$ – Nike Dattani Jun 11 at 19:38
8
$\begingroup$

Quantum Mobile Virtual Machine

Disclaimer : Not exactly a Linux Distibution but a Virtual Machine (VM)

Quantum Mobile is a virtual machine for computational materials science.The VM is based on Ubuntu 18.04.4 LTS with VirtualBox 6.1.4.

It comes with simulation codes - BigDFT, cp2k, fleur, Quantum ESPRESSO, Siesta, Wannier90, yambo, together with AiiDA and the AiiDA lab jupyter environment.

Tools included -

  • Atomistic - xcrysden, jmol, cif2cell, ase, pymatgen, seekpath, spglib, pycifrw

  • Visualization - grace, gnuplot, matplotlib, bokeh, jupyter

  • Simulation environment - slurm, OpenMPI, FFT/BLAS/LAPACK, gcc, gfortran, singularity

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Nice addition. Does it target periodic systems, or can it be used with molecular systems as well? $\endgroup$ – ksousa Jun 30 at 14:43
  • $\begingroup$ I think it's more targeted for learning purposes. Should be useful for molecular systems as well $\endgroup$ – Thomas Jun 30 at 15:35
4
$\begingroup$

Clear Linux

I think that if you need performance Clear Linux is the best choice. It's a distribution created by Intel but give also the best performance also with PC with AMD CPU respect to other distributions. Look at phoronix.com for benchmark.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.