2021-03-18 Gromacs on FAH Meeting notes

Owned by David Hahn, created
Last updated: Apr 12, 2021 by Lorenzo D'Amore
3 min read

Date

Mar 18, 2021

Participants

  • @David Hahn

  • @Lorenzo D'Amore

  • @David Dotson

  • Yunhui Ge

  • Lucie Delmotte

  • Vincent Voelz

  • Bill Swope

Goals

  • Requirements discussion:

    • What input files and directory structure is needed to run Gromacs calculations most efficiently and smoothy on FAH?

    • Any additional files/information describing our work units?

  • Where to submit/who grants permission?

    • who hosts the work server?

  • Anything else.

Discussion topics

Presenter

Notes

Presenter

Notes

David H.

  • Free energy calculations expensive; OpenFF has been using gromacs to benchmark on protein ligand systems using free energy calculations with pmx (from Bert de Groot’s lab)

    • limited resources locally, so Folding@Home would be a major improvement to our iterations

  • What would be the requirements for getting up and running with FAH?

Vincent V.

  • Which is the current state of the interface of F@H right now?

  • David D. would be my first experience with F@H for me, needed a workserver for connecting the clients to execute the work unit.

  • David D. Hadn’t discuss with John this yet.

  • Any plans for the Mobley lab for F@H? Getting a server shouldn’t be that hard

  • David D. I could coordinate with John and David M. for this

  • David H. Does the server distribute the work over F@H?

  • Lucie:

    • just need an XML spec indicating what to run; in principle can execute anything you can run under normal single-node gromacs execution

  • Vince:

    • there is some etiquette that you must respect; need to benchmark work units (WUs) to ensure your simulations don’t crash volunteers' machines, consume too much memory, determine how many points each WU is worth, etc.

    • because the WUs are designed to go to someone’s computer, finish meaningful work in a day or two, and then be sendable back

    • then making sure that you can turn around the next WU from one whose results are received

    • Lucie: doc for setting up the work server and F@H slack

Lucie

  • Access grants needed for Lorenzo and David D:

    • fah-docs

    • tester slack

    • beta slack

    • Beta testing forum: https://foldingforum.org/

      • get specific access to Beta testing of Projects and FahCores members only forum

Vincent

  • Which is the science plan exactly? Benchmarking at each FF update?

    • David H. Running a benchmark for each FF release would be ideal

      • about 20 targets in the full benchmark set

    • Smaller sets could be tested for sub-releases

  • How much simulation time is planned?

    • David H. benchmark protocol on slides

  • In F@H more demanded resources are cpus rather than gpus

    • Lucie: gromacs can run in either GPU or CPU mode; only code that can run purely on CPU, but a dearth of jobs for CPUs at the moment

    • David H. Janssen protocol is with GPUs?

  • need to consider data storage for results

  • VV: Latest gromacs is 2021: believe it’s in testing now; might have to officially release a new core to support it

  • VV: Really interested in using expanded ensemble methods! Problem is:

    • in order to get them to work, if WU isn’t long enough, histogram won’t get flat enough, so convergence never achieved

      • there is an effort to address this with an adaptive method for achieving convergence

  • Lucie: are these capabilities already in Gromacs 2021?

    • not yet, but being worked on; will need to be in next major release of Gromacs, then new core released

  • DH: we are using CPUs, so good fit for FAH demand

Lucie

  • Where does pmx fit into the workflow?

    • DH: used for generating the TPR, preparing TOP and other prep files

    • I usually use a bashscript

    • Lucy: you will have to translate it into pmx

    • VV: this approach wouldn’t work well with F@H but we can work on it. From our exp short WU wouldn’t work really well

    • LD: could put constraints on machine types for the fast simulations to run on slower machines

    • VV: possible to make equilibrium runs happen in simulation under gromacs? Stick around at a particular lambda window in your extended ensemble?

      • also perhaps run these short equilibrium runs on the work server itself if they really are short and fast

  • David D. is that any other computer intensive part?

    • DH just equilibration, minimization

Vincent

  • need an emailable point of contact for the work server and managing issues, crises

    • also an interface with the FAH org, goings-ons, issues

  • LD: for each project, need to write a short, lay-person summary for what it is, what the aims are, purpose; relevance to medical or pharmacological is valuable

    • DH good would be a dashboard for showing results, summaries

  • VV: putting some effort into PR is important; a blog post on the FAH blog valuable for communicating the importance of the effort, where it fits into larger efforts

 

 

Action items

@David Dotson will ask John and David M. about a FAH work server
@David Dotson will pursue access grants for @Lorenzo D'Amore , @David Dotson , optionally for @David Hahn
@Lorenzo D'Amore will work with @David Hahn to write a brief, one-paragraph summary of the project for non-experts
@David Hahn will draft a protocol doc giving the existing approach, given the starting points for his benchmark set; @Lorenzo D'Amore will be the prime consumer, but will also be needed for creating the FAH protocol
@David Hahn will examine PLBenchmarks for necessary updates to meet needs of protocol; might reduce number of targets
@Lorenzo D'Amore and @David Hahn discuss the scientific aims with @David Mobley

Decisions