2020-08-28 QCA Submission Meeting notes

Folding@Home?

Ben

• Ben: if we can do compression of output well, might be able to support

• David: I’ll start the PR for INCOMPLETE restarts

• David: I’m interested in Folding@Home for QCA if:

  • the business case is there (we get something we don’t have already, but really want, and have the resources to spend developing that compute pathway)

  • it fits into the stated mission of Folding@Home

Local testing for server development

Ben

• Ben: has a 2GB dump he can share that includes a partial DB dump

  • David: interested in using this to stand up a local testing env

QCSubmit ignore_errors

Josh

• Josh: can now submit with explicit ignoring of errors to support submissions that we know won’t all complete successfully

  • valuable for adding supplemental compute post-initial-submission

• David: will review #136 immediately after call; ready to merge then!

Docker deployments now up

David

• Now have automated Docker image builds to DockerHub from prod env changes

• Jeff: Galileo needs a way to pin to a particular version

  • David: Our prod images get pushed with a tag corresponding to the git commit hash they were generated from; latest tag always points to most recent (so it moves)

Galileo

Josh, Jeff

• Josh: need Docker image for server, worker separately

• Jeff: Mainly want to pursue this pathway for folks that want to run the bespoke workflow entirely on a stack they control

• David, Jeff: let’s put the Dockerfile for the server, specialized workers in the bespoke-workflow repo; images pushed to openforcefield Dockerhub org

New compute

Trevor

• New workers spun up on two new resources

  • ANI and MM : 1000 eachmapo

• David: Perhaps rename prp slack channel to fractal-compute for current status?

• Jeff: Should try and be two-deep on each machine if possible; need a contact besides John for Lilac

Precomputed datasets

Jeff, Ben, Josh

• Jeff: are there paths for precomputed data to get into QCA?

• Ben: yes, we’re doing that for example with the ANI folks

Infrastructure Roadmap

Jeff

PCM

Josh

• PCM should be doable with a small change in the Psi4 python layer; spoke with Lori for additions

  • psi4#1986

ESP

Josh

• ESPs - returning the density is a small change in Psi4

  • getting back a working wavefunction from the stored coefficients is not something that is currently implemented

  • On Lori’s roadmap psi4#1987

Disaccharides

David, Josh, Jeff

• Jeff: disacharrides submission itself not high priority/urgent

  • however, with the protein submission (which we only got PDBs for), there are suspected issues with this approach that are fundamental

  • can’t reliably generate CHMILES

  • anticipating issues with the protein dataset we’d like to avoid with the pathway we are pursuing with the disaccharide dataset

Dataset standards

Trevor, Jeff, David

• Trevor: adding more detailed field requirements for submissions

  • David: this is the policy doc we use to express the world we want for our datasets

    • this is the starting point for discussions around implementation or procedure

• Trevor: Major version tied to STANDARDS version

  • minor version is changes to the dataset

  • e.g. validation of errors depending on version

• Control new names - we should be able to do this

• Major version tied to STANDARDS version

• Need to communicate molecule info in the name

  • via version some other aspect

• Jeff: use a date for the name?