offline installation of CMI

[He Lin]

Dear All，

   Does anyone have experience in installing CMI on offline machine? Or is this feasible in reality?  We want to run CMI on some larger machine for some calculation, unfortunately the machine itself is forced offline, so we can't use conda to do it.

   Best,

                   he 

   

[Simon Billinge]

We haven't done it, but the approach we would probably take is to use conda-pack which allows you to pack up a complete conda environment and then you can copy it onto the computer that is offline and unpack it.

S

--
You received this message because you are subscribed to the Google Groups "diffpy-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to diffpy-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/diffpy-users/42faf0ed-41cf-45b8-8c1f-6f390748d925n%40googlegroups.com.

--

Simon Billinge

Professor, Columbia University

Physicist, Brookhaven National Laboratory

[He Lin]

Hi, Simon:

    That sounds great, do you need more info from our machine type/setup?

[He Lin]

The calculation seems take time for large atom clusters, so we are thinking about  parallelizing it,

 maybe we can give this a try? 
   he

[Simon Billinge]

Hi He,

We don't have any experience with the conda-pack for diffpy-cmi, which includes c++ code.  I imagine that it has to be built on an identical system (i.e., same operating system, not same HPC system) for it to be unpacked and run out of the box, but as I say, we have never tried.  If you do try, please let me know and we can think about sharing code this way.  Another thing to think about of course is using Docker.  Again, we have not focussed on this ourselves.

Pavol did get a parallel version of the pair iterator working in diffpy.sreal, but as I recall was not blown away by the speed-up, but it will depend on what you are trying to do.  For crystalline materials where the outer-loop is a sum over atoms in the unit cell, and the unit cell tends to be small,  you can parallelize the outer loop but the time is bound by the inner loop which takes the longest, especially if you are computing to reasonably high values of r.  I think this is why we didn't push through and finish that work.  But if you are doing 'big box' modeling where the outer loop is also order N (the number of atoms in the box) the speedup could be significant.   I can ask Pavol what happened to his code.    Alan Coehlo speeded things up a lot by removing the peak-broadening convolution from the innermost loop.  I have wanted to code this up in diffpy.srreal for a long time, but never quite found the right person to do it.   This might be a good place to start if you are interested in contributing to the diffpy project.  

For everyone on the thread, Diffpy is an open source community project and we welcome contributions in the form of pull-requests.  Anyone who wants to join the party, we can help you get started with the workflow....

S

To view this discussion on the web visit https://groups.google.com/d/msgid/diffpy-users/da05bee4-0268-499b-88a6-5704f90dc01fn%40googlegroups.com.

[He Lin]

Thanks, Simon, I will write to you about more details on this then.

     he

[Mark Dean]

Dear He,

I had the same problem installing other packages on an ssh-accessible server without internet access. I found that using a reverse proxy is by far the best solution

https://laptrinhx.com/securely-install-jupyter-python-without-internet-connection-in-4-easy-steps-3611876677/

Best,

Mark

[Simon Billinge]

Thanks Mark, this is very helpful.  We never tried this.

S

To view this discussion on the web visit https://groups.google.com/d/msgid/diffpy-users/aec8e016-7fbd-4017-b7c8-5d52faffc7f7n%40googlegroups.com.

[He Lin]

Dear Mark:

    Thank you very much, I will give this a try then,

     Best,

                he