GPS organization report

Massimo Gisonni, Fernando Santana Pacheco, Mandana Safari

Setting up the project

For the GitHub setup we created an ad-hoc organization. This was done to circumvent the annoyance of not being able of forking one's own repository. We implemented cmake workflow actions as well as main branch protection, requiring approval by all members of the organization before any merge. We took great advantage of in person meetings for planning ahead.

Work subdivision

• Mandana took care of Force optimization and implementation of Newton's 3rd law. Her report can be found at https://github.com/Project-MD-GPS/ljmd/blob/optim%2Bmorse/reports/report_opt.md
• Fernando took care of OpenMP implementation. His report can be found at https://github.com/Project-MD-GPS/ljmd/blob/openmp/report-openmp.md
• Massimo took care of MPI implementation. His report can be found at https://github.com/MGisonni/ljmd/blob/Final_commit/reports/Individual%20Task%20Report.md

Benchmark of results

Specifics

The simulation were run on 1,3,6,9,12 nodes of the regular1 partition in Ulysses cluster. The following modules were loaded

module load cmake/3.15.4
module load intel/2021.1
module load openmpi3

• The Benchmark is executed on the provided argon_108.*, argon_2916.* and argon_78732.* files
• All simulations are executed on full nodes of regular1 partition in Ulysses. The nodes have the following specifics
  • Model: IBM System x(R) iDataPlex(R) dx360 M4 Server
  • CPU: 2 x Intel(R) Xeon(R) E5-2680 v2 @ 2.80GHz
  • 2 CPUs with 10C/20T
  • RAM: 40 GiB
• compiler is GNU 8.3.0
• flags are -O3 -ffast-math -fomit-frame-pointer

Hybrid OpenMP+MPI implementation

The final version comes equipped with Force optimization, Newton's 3rd law implementation and hybrid OpenMP and MPI implementation. The result is a merge of each individual contribution, in particular: OpenMP is implemented using a private copy of the forces array; MPI is implemented with Balanced chunks. We refer to the individual reports for further details.

Hybrid	argon_108	argon_2916	argon_78732
3 nodes	0.045s	2.720s	16.695s
6 nodes	0.053s	1.998s	8.623s
9 nodes	0.058s	1.657s	6.082s
12 nodes	0.067s	1.568s	4.867s