2022-10-04 QC Meeting notes

Owned by Pavan Behara, created
Last updated: Oct 04, 2022 by Chapin Cavender
2 min read

 Date

Oct 4, 2022

 Participants

  • @Pavan Behara

  • @Chapin Cavender

  • @Jeffrey Wagner

  • Ben Pritchard

  • @David Dotson

 Discussion topics

Item

Notes

Item

Notes

Updates from MolSSI

  • BP – No update from me. Next branch is still going, plans are mostly on track.

  • BP – We had a hard drive issue, filled up by the write-ahead log from postgres. I think that happens if there’s a lot of changes happening.

  • BP – There was also a server overload issue, I think this was because there were managers with no work and they’d overwhelm the server with queries

Infrastructure advances

  • JW – Recharge is not updated for OFFTK 0.11, it’s on our backlog but isn’t high priority.

Throughput status

  • OpenFF Protein Capped 3-mer Backbones v1.0

    • Opts: 277295 → 289670 → 293477

    • TDs: 6 → 12 → 16

    • DD – Is this proceeding at an appropriate rate for you, CC?

      • CC – yes, these aren’t blocking

      • JW – I’ll ask LPW to cut the GeomeTRIC reelase.

  • Moved all spice sets with openff-default to end of life. Basically all failures now are mbiis failures. I mentioned this issue in the psi4 repo, trying to bring this to their attention.

  • JW – It looks like the queue is drying up, have we scaled down compute delployment?

    • PB – Yes, DD turned down PRP.

    • DD – I’ve also scaled down Lilac. That only has 64 workers, and PRP and MPI are fully turned off.

    • BP – Could you reduce the manager update frequency (actually INCREASE the period)? I may be able to reduce the heartbeat frequency on the server, but that’s already 30 minutes so I doubt it’s a problem.

User questions/issues

  • PB: Any chance spice sets can be updated on QCA’s machine learning datasets, just for visibility and not a needed feature.

    • BP – Thinking about this, there’s a few angles.

      • You should be able to just tag the set “machine learning” and it will get picked up by the website

      • The harder part is creating these HDF5 files, which isn’t automated. I’ve also been wanting to deprecate the HDF5 part. It’s such a manual process that I’ve kept from adding new sets. And all the HDF5s are custom.

    • BP – So the 80/20 solution might just be tagging the dataset and us agreeing to leave out the HDF5. Another software scientist, Sina, is taking over the ML portion and so they may want to make the call here.

    • PB – The dataset already comes with a downloader script from our end, so that should be an OK replacement for HDF5.

    • BP – You can look at the TensorMol Water Clusters dataset for an example of a dataset with no HDF5

    • BP – We also have a new postdoc who may be able to do the HDF5 curation, but they’re not ready yet. So I may put you in contact later.

  • PB – Do we have any dimer interaction datasets on QCA?

    • DD – Don’t we have a SPICE dimer dataset?

    • CC – I think there are some pepconf dimers.

    • JW – There’s a terrible hack in QCSubmit that allows for multiple molecule submission, so I suspect that was put in place for a dimer dataset

      • PB – That could have been solvated amino acids and DES dimers

    • DD – What kind of dataset are you looking for?

    • PB – I’m looking for dimer datasets to follow up on a paper I’d read

    • BP – COMP6, S66,

Science support needs

 

 Action items

 Decisions