PaperlayerView exact revision on arXiv

Source-linked AI summary

FabSim: facilitating computational research through automation on large-scale and distributed e-infrastructures

Derek Groen, Agastya Bhati, James Suter, James Hetherington, Stefan Zasada, Peter Coveney

arXiv:1512.02194v1cs.DCphysics.comp-ph

TL;DR

Researchers must manage many computational models and runs, making routine tasks and reproducibility demanding. FabSim automates these accompanying tasks and organizes run information; its use across three research areas demonstrates its flexibility, though repeated computations can still differ when software builds or resources change.

  • Problem

    Ensemble and multiscale research requires careful management of many models and runs, and mistakes can produce incorrect results or insufficient information for reproduction.

    Source: §1 · p. 3

  • Method

    FabSim provides a toolkit and best practices for automating computational research tasks and curating inputs, outputs, configurations, and execution instances.

    Source: §1 · pp. 3–4

  • Results

    FabSim has been applied to bloodflow, materials modelling, and binding-affinity calculations, with domain-specific workflows including multiscale simulation, potential parameterization, and ensemble computations.

    Source: §5 · pp. 20–21

  • Takeaways & Limitations

    Across these use cases, FabSim assists run-data curation by recording job specifications and environment variables, supporting the authors’ conclusion that it benefits computational research generically.

    Source: §5 · pp. 20–21

    Repeated computations may still produce different results when remote modules are recompiled with different settings or compilers, or when different resources are used.

    Source: §3.4 · p. 11

Abstract

from arXiv · show

We present FabSim, a toolkit developed to simplify a range of computational tasks for researchers in diverse disciplines. FabSim is flexible, adaptable, and allows users to perform a wide range of tasks with ease. It also provides a systematic way to automate the use of resourcess, including HPC and distributed resources, and to make tasks easier to repeat by recording contextual information. To demonstrate this, we present three use cases where FabSim has enhanced our research productivity. These include simulating cerebrovascular bloodflow, modelling clay-polymer nanocomposites across multiple scales, and calculating ligand-protein binding affinities.

PROGRAM SUMMARY/NEW VERSION PROGRAM SUMMARY

FabSim is a toolkit for simplifying and automating computational research tasks, particularly on remote resources. Its command-line interface assumes users have some scripting knowledge, and the paper reports applications across bloodflow, nanomaterials, and ligand-protein binding.

PROGRAM SUMMARY/NEW VERSION PROGRAM SUMMARY

1. Introduction

Ensemble and multiscale research increasingly requires careful management of many model executions, while routine accompanying tasks consume researchers’ attention. FabSim is presented as a customizable toolkit and set of practices to automate these tasks and organize computational runs.

1. Introduction

  • Ensemble and multiscale computations require researchers to construct, execute, validate, analyze, and curate multiple models and runs, where mistakes can undermine results or reproducibility.Source: §1 · p. 3
  • Routine work accompanying computations consumes researcher attention and can limit research pace to the person-hours invested.Source: §1 · p. 3
  • FabSim combines software tools and best practices to automate research tasks and maintain a computational environment that is easy to use and modify.Source: §1 · p. 4
  • FabSim reduces administrative tasks to single-line commands and systematically curates inputs, outputs, configurations, and execution instances.Source: §1 · p. 4

2. Related work

FabSim overlaps with grid middleware and workflow tools but emphasizes an extensible command-line environment for computational scientists. Its use of SSH avoids requiring an additional heavyweight middleware installation on accessed resources.

2. Related work

  • Unlike monolithic grid middleware such as Globus, Unicore, and gLite, FabSim targets experienced computational scientists with an extensible command-line interface.Source: §2 · p. 4
  • FabSim uses SSH and does not require an additional heavyweight middleware stack on accessed resources, supporting use on HPC systems that support SSH.Source: §2 · p. 4
  • Longbow offers shorthand commands for distributed molecular-dynamics runs, but does not explicitly support multi-step workflows or code compilation and deployment.Source: §2 · pp. 4–5
  • Workflow and coupling tools can complement FabSim, while its stated strength is simplifying development of new workflows and computational approaches as well as executing existing ones.Source: §2 · p. 5

3. Overview of FabSim

FabSim pairs customizable software with best practices to simplify and systematize computational research tasks. Its modules support remote execution, curation, reproducibility, and SSH-based security.

3. Overview of FabSim

  • FabSim pairs customizable software with best practices to help researchers perform computational tasks more simply, quickly, and systematically.Its central purpose is to save time for computational researchers.Source: §3 · pp. 5–6

3.1. The software

  • FabSim is written in Python, requires no administrative privileges to install, and uses Fabric for remote-machine access and YaML for configuration data.All modules are installed on the user's local workstation.Source: §3.1 · p. 6, Figure 1 · p. 6

3.1.1. Using FabSim

  • FabSim features are called from the terminal using a command structure, and commonly used commands can take parameters or use machine-specific defaults.Domain-specific commands are also available and can be incorporated into more complex user-made commands.Source: §3.1.1 · p. 7

3.1.2. Customizing FabSim

  • FabSim uses YAML variables and templates to customize scripts, combining application and scheduler templates without requiring a separate template for each combination.Customization can be machine-specific, user-specific, or both; features are organized across general-purpose, machine-specific, domain-specific, and problem-specific scopes.Source: §3.1.2 · pp. 7, 9

3.2. Best practices

  • FabSim's best practices separate machine and user configuration and keep general-purpose, domain-specific, and problem-specific features in distinct locations.Machine defaults belong in machines.yml, while each user's machine settings are stored locally in machines_user.yml.Source: §3.2 · p. 9

3.3. Remote execution

  • FabSim supports one-line remote computations, can stage local input files to remote resources, and can combine remote jobs with local processing tasks.Users can also chain interdependent jobs in a single FabSim command.Source: §3.3 · pp. 9, 11

3.4. Provenance, curation, reproducibility

  • FabSim records execution context, remote environment variables, and submission scripts to help users inspect configurations and repeat computations.Repeated runs can still differ when remote modules are recompiled with different settings or compilers, or when different resources are used.Source: §3.4 · p. 11

3.5. Security

  • FabSim uses SSH-based authentication with public/private keys, but users must manage their own key setup on target resources.Key management may add setup overhead, though the authors note most users will already have configured SSH keys for resources they access.Source: §3.5 · p. 12

4. Exemplar FabSim use cases

FabSim supports research workflows across cerebrovascular bloodflow, clay-polymer nanocomposites, and ligand-protein binding, including multi-simulation execution and analysis. These examples show how domain adaptations automate work while supporting studies of bloodflow, coarse-grained materials models, and binding-affinity estimates.

4. Exemplar FabSim use cases

  • The use cases span three scientific domains, with FabSim adapted to run and analyze workflows for bloodflow, multiscale materials modelling, and ligand-protein binding.The examples include HemeLB simulations, FabMD coarse-graining, and ensemble molecular dynamics for binding-affinity calculations.Source: §4 · p. 13, §4.1 · p. 13, §4.2 · p. 16

4.1. Cerebrovascular bloodflow

  • FabSim and HemeLB support reliable predictions of bloodflow properties in cerebral arteries and mouse retinas, while enabling simulation sets for scalability and accuracy studies.FabHemeLB also automates ensemble multiscale simulations and performance tests; the workflow compares wall shear stresses across exercise intensities.Source: §4.1 · pp. 13–15

4.2. Multiscale modelling of clay-polymer nanocomposite materials

  • FabMD automates coarse-grained parameterization of clay-polymer systems using IBI or, when IBI is ineffective, the more computationally expensive but robust PMF method.IBI adjusts pair potentials iteratively to match all-atom radial distribution functions; each PMF iteration uses approximately 20–40 constrained-distance simulations.Source: §4.2 · p. 16

4.3. Calculating ligand-protein binding affinities from ensemble molecular dynamics

  • Repeated binding-affinity protocols produce a Gaussian distribution whose ensemble average estimates the theoretical affinity, with bootstrapping providing tight error bounds.FabBioMD automates distributed workflow steps, and a 50-replica study compared predictions for 12 peptide sequences bound to an MHC allele with experimental affinities.Source: §4.3 · pp. 18–20

5. Conclusion

FabSim reduces the complexity of computational research and supports reproducibility across bloodflow, materials modelling, and binding-affinity calculations. The three use cases demonstrate its flexibility across research domains.

5. Conclusion

  • FabSim’s three use cases in bloodflow, materials modelling, and binding-affinity calculation provide evidence that it benefits computational research on a generic level.The authors describe FabSim as a way to reduce the complexity of computational research tasks and improve reproducibility.Source: §5 · p. 20
  • FabSim supports systematic benchmarking, ensembles of multiscale simulations, and simpler deployment of HemeLB on remote machines for brain bloodflow research.Source: §5 · p. 20
  • 7,200 simulation and 4,800 free-energy cores are needed for a single-trajectory calculation; three concurrent trajectories require 19,200 and 14,400, respectively.The workflow takes approximately 9 hours when sufficient resources are available; concurrent peptide calculations multiply resource needs by the number of peptides.Source: Figure 7 · p. 21
  • In nanomaterials, FabSim automates iterative coarse-grained-potential parameterization and chemically specific modelling of layered-composite self-assembly; for ligand-protein binding, it automates ensemble calculations and data curation.Run data records job specifications and environment variables; the binding workflow is described as helping control uncertainties and produce reproducible results.Source: §5 · p. 21
  • Figure 8 presents Gaussian frequency distributions from 50 simulations total across two peptide-binding-affinity calculations (WIK-TISKRM and AAAKTPVIV), alongside a comparison of calculated affinities with experimental results.Source: Figure 8 · p. 22
  • FabSim’s applications span three domains in which the authors report publishing research findings in leading journals; the code is released under LGPL version 3.The authors hope the release will encourage researchers to automate tedious tasks and free more effort for advancing science.Source: §5 · p. 22
Loading 1512.02194v1…