2020-03-19 QCFractal Meeting notes

Updates from MolSSI

• DGS – Server failure this weekend, we’ve updated to a newer, bigger box with MUCH more memory

• DGS – We’ve been working on web apps and papers the last month

Queue/worker status

DGS – Green planet is only fielding one manager with 16 cores – @Trevor Gokey should check into this after finals

DGS – 160/400 torsiondrives are complete in five days. The first 100 went really fast, then these 60 are a bit slower. So molecule size makes a big difference.

JRW – We’re going to submit 2 or 3x the current amount of torsion drives to the queue, as well as ~2000 geometry optimizations, and hope for them to be finished by April 20. Is that reasonable, and if so, what steps can we take to ensure they finish?

DGS – You may be able to complete that on own OpenFF resources alone. Each torsiondrive is several hundred geometry optimizations, so we could finish ~2000 geometry optimizations in about one day. MolSSI can contribute 500 concurrent cores, but will want us to “pay them back” approximately equally in the future.

JRW – That sounds good to us. We made no statement to PRP about what we’re using their resources for, so compensating MolSSI for this time with PRP should be fine.

Current Issues

HJ sees 7 failures in existing torsiondrives – 2703602 for example

DGS – It looks like the error wasn’t captured when it happened, and this job was originally run in 2019 on an older version with less logging. I’ve restarted it.

HJ – What’s the best way to approach these in the future?

DGS – Check for convergence errors, since restarting won’t help. Resource errors should be resubmitted automatically. But for other things, just collect every case of them that you find and send them to me. We have the automation to successfully run 95% of torsiondrive jobs, but the remaining 5% have such diverse failure modes that we haven’t automated handling for all of them.