Velocity Auto-Correlation Function (VACF)

Serial and parallel implementation of Velocity Auto-Correlation Function in C with MPICH for the paper titled "Calculation of the Velocity Auto-Correlation Function Using Parallel Algorithms" by Parth Vipul Shah, Shubhadeep Nag, Debnath Pal, Subramanian Yashonath. Submitted for review.

Introduction

Abstract: WIP

TODO: Update

This repository contains all the resources for the reproduction of our results which are detailed in our paper. The data required can be generated using LAMMPS.

Setup

To run these algorithms on a cluster, you will need MPICH, an implementation of the Message Passing Interface (MPI). To generate data, you will need LAMMPS.

MPICH

MPICH can be installed from here or a package manager like apt. Documentation is available here. Once installed, run hello_world.c to verify installation.

Compile and run with:

$mpicc -o test hello_world.c -std=c99

$mpiexec.hydra -n 4 ./test

More information can be found on the official MPICH documentation.

new_script.sh is a sample script provided for your reference. We used this to submit jobs to our PBS queue running on a 120 node cluster with Intel Xeon E5-2670 (Haswell) CPUs.

LAMMPS

LAMMPS can be downloaded from here. Unpack the same and run. The Manual is available here.

Run a simulation using:

$(lammps executable) < (lammps input script)

Usage

All programs take 3 arguments during execution. The massively parallel implementations take one additional parameter.

Flags
---------------------------
-p int,int,int details the start,stop, step timesteps
-a int details the number of particles in the system
-i int details the number of correlations that need to be calculated
---------------------------
[-bp int] (optional) details the number of particles in one batch for the massively parallel algorithms

Example

To compile and run Mpar1.3.1.c on 3 nodes with 24 processors each:

$mpicc -o vacf.Mpar1.3.1 Mpar1.3.1.c -std=c99

$time mpiexec.hydra -np 72 -genvall -ppn 24 ./vacf.Mpar1.3.1 -p 50000,9857960,10 -a 6912 -i 1000 -bp 144 > out.data 2>&1

Organization

This project has multiple directories. Here is a brief description of each. Please navigate to the specific directories for more information.

algos

This directory contains all the algorithms, serial and parallel.
lammps

This directory contains the details for the data generation using LAMMPS, a widely used parallel software for Molecular Dynamic simulations.
timing

This directory contains CSV's with timing data. For more (and complete) timing information, visit this repository or refer to the tables in our paper.
utils

This directory contains the utility functions associated with this project. They are python scripts.

Citing

Cite this work as: WIP