Skip to main content

mdbenchmark gromacs simulations

Project description

https://travis-ci.org/bio-phys/MDBenchmark.svg?branch=master https://codecov.io/gh/bio-phys/MDBenchmark/branch/master/graph/badge.svg

Quickly generate, start and analyze benchmarks for GROMACS simulations. Below is an example of the scaling plots you can generate quickly with MDBenchmark.

runtimes.png

Installation

You can install MDBenchmark via pip:

pip install mdbenchmark

If you are using conda:

conda install -c conda-forge mdbenchmark

Usage

After installation, the mdbenchmark command should be available to you globally. If you have installed the package in a virtual environment, make sure to activate that first!

Generate 10 benchmarks for our system with the GROMACS module gromacs/5.1.4-plumed2.3.

mdbenchmark generate --name protein --module 5.1.4-plumed2.3 --max-nodes 10

The naming of the GROMACS module assumes that you refer to the GROMACS module you are using. On our clusters this is always gromacs/VERSION.

To run benchmarks on GPUs add the --gpu flag:

mdbenchmark generate --name protein --module 5.1.4-plumed2.3 --max-nodes 10 --gpu

You can also create benchmarks for different versions off GROMACS:

mdbenchmark generate --name protein --module 5.1.4-plumed2.3 --module 2016.4-plumed2.3 --max-nodes 10 --gpu

After you generated all the benchmarks, you can start them at once:

mdbenchmark start

When the benchmark simulations have finished, you can run the analysis to produce a CSV output file or a plot for direct usage (via --plot option).

mdbenchmark analyze

Defining Host Templates

It is possible to define your own host templates in addition to the ones shipped with the package. A template file should have the same filename as the UNIX command hostname returns to be detected automatically. You can also specify the exact template you want to use with --host.

Here is an example template from for the MPG cluster hydra.

# @ shell=/bin/bash
#
# @ error = {{ name }}.err.$(jobid)
# @ output = {{ name }}.out.$(jobid)
# @ job_type = parallel
# @ node_usage = not_shared
# @ node = {{ n_nodes }}
# @ tasks_per_node = 20
{%- if gpu %}
# @ requirements = (Feature=="gpu")
{%- endif %}
# @ resources = ConsumableCpus(1)
# @ network.MPI = sn_all,not_shared,us
# @ wall_clock_limit = {{ formatted_time }}
# @ queue

module purge
module load {{ module }}

# run {{ module }} for {{ time }} minutes
poe gmx_mpi mdrun -deffnm {{ name }} -maxh {{ time / 60 }}

MDBenchmark provides the following values to a template

  • name: name of benchmark

  • gpu: boolean if GPUs are requested

  • module: module to load

  • n_nodes: number of nodes to run on

  • time: program run-time in minutes

  • formatted_time: queuing runtime in human readable format (HH:MM:SS)

To ensure correct termination of jobs formatted_time is 5 minutes longer then time.

MDBenchmark searches for user templates in the xdg config folders defined by XDG_CONFIG_HOME and XDG_CONFIG_DIRS which by default are $HOME/.config/MDBenchmark and /etc/xdg/MDBenchmark respectively. If the variable MDBENCHMARK_TEMPLATES is set, the script will also search in that directory.

MDBenchmark will first search in XDG_CONFIG_HOME and XDG_CONFIG_DIRS for a suitable template file. This means it is possible to overwrite system-wide installed templates or templates shipped with the package.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

mdbenchmark-1.0.1.tar.gz (35.9 kB view hashes)

Uploaded Source

Built Distribution

mdbenchmark-1.0.1-py2.py3-none-any.whl (19.2 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page