2023-01-17 QC Meeting notes

 Date

Jan 17, 2023

 Participants

  • @Pavan Behara

  • @Jeffrey Wagner

  • @David Dotson

  • BenjaminPritchard

Discussion topics

Item

Notes

Item

Notes

Updates from MolSSI

  • BP: Meetings with collaborators keeps me busy.

  • BP: A few things to do on next branch, related to how services work.

  • BP: Keeping tabs on dataset discussion about inconsistent results

    • PB – This is a Psi4 issue, not QCF. You’ll get sucked in if you respond, recommend not

    • BP – I was thinking about allowing version restrictions in tasks

    • PB – Could we put something like that in …

    • BP – Yes, that’s what I was thinking. But it would take some time to add. Wouldn’t go in qcschema, instead would be in fractal/portal. Also would need to be a very fast check

    • DD – Operationally, it would also make compute management more complex. I’d be against this.

    • PB – This would be a stopgap on running with old versions.

    • DD –

    • BP – This sort of thing could be handled in the dataset tags.

    • DD – True, but tags aren’t really standardized. They’re ephemeral and operators have to make up meanings for them. Maybe this is worth implementing on a more formal level.

    • BP – That’ll be hard. We’d basically need to implement version comparison. It’d be easiest if we only looked at version equality, without < or >. Would want this to be postgres side. Would be more general than psi4. In the next branch, managers tell server which programs they have available. Would like to handle this on a database level

    • JW – Since next will allow us to delete results, maybe it would be cheaper to occasionally have to delete and rerun some results?

    • BP – Yeah, next will allow deletion. And it will also let us resubmit calcs.

    • DD – Manual reruns may be easier.

  • BP – I’ll send out next branch docs.

  • BP – Actions on SPICE dataset?

    • JW: Discussed with lead team - I don’t see us as owning SPICE set. If JC is requiring additional work please ensure that he’s being billed for the time

    • DD – PB, are you planning to continue work here?

    • PB – Unsure

    • JW – PB, you should check with DM whether he wants you to work on this.

    • PB – YW uses SPICE for espaloma training, and LW may use it for charge training.

    • PB – Should we rerun the bad energy calcs?

    • JW – Compute time is free, human time is expensive. So if we can do this in under 3 human hours then let’s do it

    • BP – How do we trigger a recomputation of these jobs?

    • DD – Are they all in one dataset?

    • PB – No

    • DD – So this would kinda be like rerunning the whole SPICE dataset again.

    • PB – We’d resubmit but many would match the original jobs. Only the subset that we need to rerun would be rerun (they’d be submitted with a slightly different input so they don’t match with the previous submission)

    • DD – That sounds like a lot of detailed work.

    • PB – probably 2 days of work.

    • DD – Yeah, I think we should wait until we can delete with the next branch.

    • JW – I think we should take no action

    • PB – Could we say we’ll do it when we have time (over 2 months or something?), or when QCF next comes out.

    • BP – This will be easy once next comes out. But I don’t need more pressure for the release. One part is that we need to have qcsubmit ready to go, and I’d pushed on this a bit a few weeks ago. Another source of uncertainty is whether the VT grant comes through and I need to acquire a new server.

    • DD – I think we should put a pin in this until next

    • PB – We could say that PE could make a new release of the dataset excluding the potentially bad entries.

    • DD – That makes a lot of sense - QCA isn’t supposed to be a system of record, so there’s no big issue here with provenance. I think filtering to avoid the bad entries is a good way to go.

    • PB – Perfect

    • DD – I’ll tell PE and JC about this decision

  • PB – Error cycling failed on Friday, something with conda

    • …

    • DD – Maybe something with py3.8 going out of date?

    • JW – Why is it looking for libgcc on defaults?

    • DD – Could try taking defaults out of channels list.

    • PB – I’ll try these. I can’t commit to main though.

    • DD – I’ll give you admin permissions to the repo. (JW confirmed admin is appropriate)

    •  

Infrastructure advances



Throughput status

 

User issues

  • JW: CC mentioned 2D torsiondrives were incomplete and would look into the issues, and want to be complete at the time of publication.

    • DD: So, no action needed now until he reviews?

    • JW: Yeap.

    • DD: I think we should archive calculations and stop using QCA as our sole data source, since it is more of a facilitator for creating datasets.

 Action items

 Decisions