UCL-RITS / rcps-buildscripts

Scripts to automate package builds on RC Platforms
MIT License
39 stars 27 forks source link

Install Intel 2018 compiler, stop building with 2017 #171

Closed heatherkellyucl closed 6 years ago

heatherkellyucl commented 6 years ago

As well as #167 numerical issues, today's Tier2 call confirmed multiple sites having problems with the Intel 2017 compiler and MPI. (CASTEP problems and bugs in MPI collective I/O write function, Fortran unable to write file errors).

These are supposed to be fixed with 2018, and seemed to have fixed the file writing errors. Sites intending to move to 2018 as the default compiler.

Suggest we install 2018 asap and stop building things with 2017.

balston commented 6 years ago

I'm downloading the latest 2018 Update 2 bundles from the Intel Registration Centre with my UCL site login.

balston commented 6 years ago

I now have all the archives downloaded. I'm copying the Linux one on to the clusters.

balston commented 6 years ago

I've done a test install on my CentOS 6 VM to check this version licenses correctly from the UCL LM and it does.

balston commented 6 years ago

Install archive:

parallel_studio_xe_2018_update2_cluster_edition.tgz

now in:

/shared/ucl/apps/intel/installers

on Legion, Grace and Thomas. Will add to Myriad when we can scp files to it.

balston commented 6 years ago

The install archive is now also on Myriad in the same location.

owainkenwayucl commented 6 years ago

I guess we need a comms plan for migrating the default modules as we did for 2017 as well.

balston commented 6 years ago

Intel have just released update 3 so we should use this version. I'm starting to download it now ...

balston commented 6 years ago
parallel_studio_xe_2018_update3_cluster_edition.tgz

now in:

/shared/ucl/apps/intel/installers

on:

owainkenwayucl commented 6 years ago

OK - gonna take a pass at this.

owainkenwayucl commented 6 years ago
Name "PHONEHOME_SEND_USAGE_DATA" is not valid
/home/ccspapp/temporary-intel-dirs/intel-build.VyXR019Fkq/./silent_config_file.cfg has errors

Yay.

I bet there's literally no documentation for what it's changed to either.

owainkenwayucl commented 6 years ago

The documentation is a PDF because of course it is.

owainkenwayucl commented 6 years ago

Mostly dealing with Legion complexities at the moment.

owainkenwayucl commented 6 years ago

See here: https://github.com/UCL-RITS/rcps-buildscripts/commit/64c93f21049945359f3179ad6afcad7d70c0a26d

owainkenwayucl commented 6 years ago

Right, it is now installed (along with the interfaces) on all four clusters.

Now we need to see if it has the same issues as we saw in 2017.x.

heatherkellyucl commented 6 years ago

It does fix the NAMD case - boundary box no longer expanding past 59 Angstroms.

owainkenwayucl commented 6 years ago

OK, I beleive I successfully have CESM working with Intel 2018 on Myriad so I'm pushing out those changes on the other services.

ikirker commented 6 years ago

We told them we were going to switch at the end of July, so remind users on Monday to give warning.

owainkenwayucl commented 6 years ago

This is now done and we've moved so closing.