2022-01-25 QCF Meeting notes

General updates

• PB – BP is leading a workshop today and can’t attend

• DD – Storage use if growing over 30 GB/day. This will exhaust QCA’s disk space in about 2 weeks. So we’ve shut off the calculations on SPICE. BP also saw some other bulky calculations coming in, and ti looks like it was PR 259 (SPICE dipeptides dataset), which I hadn’t seen before.

  • DD – I communicated this bottleneck to PE and JC, and they asked if we could still continue on DES370K.

• DD – We’re still prioritizing OFF sets, but this is just a question of whether we can give PE and JC some compute on DES370K.

• DD - No managers on PRP, still prioritizing OpenFF jobs on other resources

• DD & PB - We should shut off SPICE pubchem set 1

• DD - SPICE dipeptide optimizations do not store wavefunctions, so that dataset should not impact storage much

• CC – The last dipeptide torsiondrive in the 1.1 set is hanging, I think because we prioritized the dipeptide torsiondrive 2.0 set

  • PB – I’ve updated priority to ensure that 1.1 gets finished.
    

Infrastructure needs/advances

• DD:

  • cutting QCEngine release today; this addresses pycpuinfo issue we've observed on PRP

  • will merge QCFractal#700; addresses cleanup issue on e.g. Lilac and elsewhere

  • will observe if both compatible with QCFractal#701, which uses latest QCElemental, QCEngine; merge

  • cut release of QCFractal

  • update all deployed environments; deploy on all compute resources

• PB – Can jobs report when they’re killed by the scheduler (in their qca-dataset-submission tracking info?)

  • DD – I don’t think so, this would cross a bunch of process boundaries, and to the manager it just looks like a keyboardinterrupt

• JW – Details about pycpuinfo issue?

  • DD – Details unknown, just that some hosts on PRP don’t give CPU info to provenance logger, and previously this would be fatal. So I’ve made it optional.