Reproducibility between versions

[jmro...@ujaen.es]

Hello,

I am currently using Sire version 2016.1. I have also asked the program installed on a cluster I do not administer and they have told me that due to glibc dependencies they can only install 14.4 (they want to continue using old glibc libraries because they say updating those libraries can destabilise the system). I am running relative free energies calculations of a protein-ligand system

So, what kind of changes have been made between these two versions? Could I mix in the same project results from both versions?

Thank you

[Christopher Woods]

  Hi,

You can mix different versions, although I generally wouldn't recommend it. Which operating system are you using, as I should be able to build a binary using an older glibc that will be compatible with your cluster.

Alternatively, building Sire from source is really straightforward. You just have to download the source 

https://github.com/michellab/Sire/archive/2016.1.0.tar.gz

and run "./compile_sire.sh" that is contained within. The only dependencies you need are a version of cmake that is greater than 2.8.11 and a version of GCC that is greater than 4.2.

  Best wishes,

  Christopher

---------------------------------------------------------
Christopher Woods
+44 (0) 7786 264562
http://chryswoods.com

--
You received this message because you are subscribed to the Google Groups "Sire Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sire-users+...@googlegroups.com.
To post to this group, send email to sire-...@googlegroups.com.
Visit this group at https://groups.google.com/group/sire-users.
For more options, visit https://groups.google.com/d/optout.

[Jose Manuel Granadino Roldán]

Good morning,

Thank you for your answer. I was waiting for the IT people to answer me. They told me that in their cluster they have  glib 2.9 on SLES 11 with kernel 2.6.27.19-5-default. It would be great if you could build a binary.

  Best

You received this message because you are subscribed to a topic in the Google Groups "Sire Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sire-users/roMjSicM4N8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to sire-users+...@googlegroups.com.

To post to this group, send email to sire-...@googlegroups.com.
Visit this group at https://groups.google.com/group/sire-users.
For more options, visit https://groups.google.com/d/optout.

[Christopher Woods]

  Hi,

Just to say that I haven't forgotten you ;-)

The latest versions of Sire use C++ 2014, so need a more modern C++ compiler. I am currently building a VM that uses an old glibc but a new C++ compiler. I will then use this to build the binary of Sire that has a glibc that supports your system. I will keep you informed of progress.

  Best wishes,

  Christopher

---------------------------------------------------------
Christopher Woods
+44 (0) 7786 264562
http://chryswoods.com

[Hannes Loeffler]

That is actually quite funny because I recently had a request for an "old" version of FESetup for an even older system: SLES10 (EOL July 31st, 2016! :-).  And from Spain too... Chris, what is the oldest glibc you would want to support?

[Christopher Woods]

  Hi Hannes,

I had hoped to move up from GLIBC 2.5 (version for SL 5, released 2006) to GLIBC 2.12 (version for SL 6, released 2010). This was because I want to start using C++ 14, and putting compliant compilers on older systems is painful. I have *nearly* got a VM image with SL 5 with clang 3.9.0 installed. It has been a pain because linking clang uses > 8 GB of memory, which has crashed virtualbox on my windows machine. I am now building the image on my macbook, as that has more memory (8 GB versus 3 GB).

I'd ideally not want to go below 2.5, but could be persuaded if there was sufficient demand and I can find a linux image in which I can build clang.

  See you next week,

  Christopher

---------------------------------------------------------
Christopher Woods
+44 (0) 7786 264562
http://chryswoods.com

[Hannes Loeffler]

Hi Chris,

I am a native Linux user so my solution is a chroot environment.  I have installed Debian etch (with debootstrap) which is glibc 2.3.6.  This setup required me to install a newer gcc (4.7 but I think that's now the minimum requirement for building 5.x), and a new version of the binutils.  But there are potential knock-on effects for other dependencies you may have.  The advantage of this setup is that it is rather minimal an can help avoid dependency creep.  A separate chroot can be used for testing.  I think typical Linux images tend to be rather "fat".

I wonder if docker could not be a more light-weight solution for you instead of booting a whole VM.  Have never tried that on Mac or Win though.

See you next week,

Hannes.

[Christopher Woods]

  Hi,

I've now created a pre-release version of Sire 2016.2 that is compiled against GLIBC 2.5. Please could you try this and see if it works on your cluster (it has exactly the same requirements as Sire 2014.4).

You can download this from here;

<https://drive.google.com/file/d/0B_KkGMZ8ACfabWtPX1hHaHU5RzA/view?usp=sharing>

This is Sire 2016.2.0. I will be releasing Sire 2016.2.1 (the proper 2016.2 release) once we have fixed some issues with OpenMM. These should not affect you, as Sire 2014.4 didn't have OpenMM, so I guess that you weren't using this functionality? (i.e. sommd).

Please let me know how you get on.

  Best wishes,

  Christopher

---------------------------------------------------------
Christopher Woods
+44 (0) 7786 264562
http://chryswoods.com

[Julien Michel]

Hi Chris, 

Unfortunately Jose is doing this to run somd, so while this should at least make it possible to compile sire, some additional tricks may be required to get OpenMM to play nicely.

Best wishes,

Julien 

--------------------------------------------------------------
Dr. Julien Michel,
Royal Society University Research Fellow
Room 263
School of Chemistry
University of Edinburgh
West Mains Road
Edinburgh, EH9 3JJ
United Kingdom
phone:   +44 (0)131 650 4797
http://www.julienmichel.net/
-------------------------------------------------------------

[Christopher Woods]

  Hi Julien,

Ok... ;-)

Strangely, OpenMM is working with this version on some machines. It is working on the machines we are using for the CCP-Biosim workshops next week, but doesn't work on the SL5 image that I compiled it (or the other SL5 machine on which I tested it).

Julien - would you mind downloading the binary in

<https://drive.google.com/file/d/0B_KkGMZ8ACfabWtPX1hHaHU5RzA/view?usp=sharing>

and unpacking it, then running $SIRE/bin/sire_test

This should run all of the tests, including the OpenMM tests. If it works, then the test will run (with one failure from a test Antonia has added but which the current code doesn't pass). If it doesn't work, then Sire will just crash with a segfault when it tries to use OpenMM.

  Thanks,

  Christopher

---------------------------------------------------------
Christopher Woods
+44 (0) 7786 264562
http://chryswoods.com

[Jose Manuel Granadino Roldán]

  Hi all,

  Thank you very much for your help. Indeed, those who asked you also about Sire are the same IT people in charge of the cluster I want to use. I just wanted tell you know that they have let me know they´ve been able to compile Sire from source (one of the 2 ways proposed). As soon as I am able to test it I will let you know if I encounter trouble with OpenMM.

  Best

<firma_español.tiff>

[JOSÉ MANUEL GRANADINO ROLDÁN]

  Good afternoon,

  I am not sure if I should start another thread, although I think this is related to what you have stated previously. While I am waiting from the people in Barcelona to fix another issue with installation (seems like numpy is not installed automatically and I need them to use conda to do it) I have tried versions 2016.1 (from the webpage) and 2016.2 (the link sent by e-mail) on a computer with Ubuntu 14.04.2 LTS (GNU/Linux 3.16.0-30-generic x86_64) and I regret to state that for both versions when trying to enter into the OpenMM part I receive the error message: 

###=======================Minimisation========================###

Running minimisation.

Starting /home/jose/sire.app/bin/somd-freenrg: number of threads equals 32

Traceback (most recent call last):

  File "/home/jose/sire.app/pkgs/sire-2016.1.0/share/Sire/scripts/somd-freenrg.py", line 142, in <module>

    OpenMMMD.runFreeNrg(params)

  File "/home/jose/sire.app/lib/python3.5/site-packages/Sire/Tools/__init__.py", line 140, in inner

    retval = func()

  File "/home/jose/sire.app/lib/python3.5/site-packages/Sire/Tools/OpenMMMD.py", line 1496, in runFreeNrg

    system = integrator.minimiseEnergy(system, minimise_tol.val, minimise_max_iter.val)

RuntimeError: There is no registered Platform called "CUDA"

  I tried to use OpenCL instead of CUDA but no success. 

 Best

  

José Manuel Granadino Roldán  Profesor Titular de Universidad  jmro...@ujaen.es  Universidad de Jaén  Departamento de Química Física y Analítica  Edificio B3, despacho 104  Campus Las Lagunillas s/n 23071 Jaén | +34 953213057 |