Skip to content

Repository of information and results for the DiRAC Extreme Scaling benchmarks.

Notifications You must be signed in to change notification settings

DiRAC-HPC/es-benchmarking-public

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DiRAC Benchmarks (Extreme Scaling)

This repository contains data and analysis for benchmarking the Grid data parallel C++ mathematical object library on the DiRAC systems.

Grid

Obtaining the source code

The source code can be obtained from the public Github repository:

All benchmarks were run using the source from the dirac-ITT branch. The typical process for checking out the code would be:

git clone https://[email protected]/paboyle/Grid
cd Grid
git checkout release/dirac-ITT

Compiling GRID

General information on compiling Grid can be found at:

with specific instructions for the benchmark at:

The build process for each of the systems we have run on can be found in the doc/ subfolder.

Running Grid benchmark

Information on running the DiRAC Grid benchamrk can be found at:

Note that the MPI task distribution across problem dimensions used at different node counts follows the following rules:

  • Distribute as evenly as possible
  • Larger numbers should be in the left-most dimension

This leads to the following distributions when using 2 MPI tasks per node (usually best on two-socket nodes):

  • 1 node (2 MPI tasks): 2.1.1.1
  • 2 nodes (4 MPI tasks): 2.2.1.1
  • 4 nodes (8 MPI tasks): 2.2.2.1
  • 8 nodes (16 MPI tasks): 2.2.2.2
  • 16 nodes (32 MPI tasks): 4.2.2.2
  • 32 nodes (64 MPI tasks): 4.4.2.2
  • etc.

The benchmark has been run on the following systems:

System Node Interconnect Combinations
DiRAC Memory Intensive (COSMA7) 2x Intel Xeon Skylake Gold 5120, 512 GiB DDR4 Mellanox EDR Intel 2018; Intel MPI 2018
DiRAC Extreme Scaling (Tesseract) 2x Intel Xeon Skylake Silver 4112, 92 GiB DDR4 Intel OPA Intel 2018; Intel MPI 2018

Output from benchmark runs can be found in the results/ subdirectory.

Evaluating benchmark performance

Information on evaluating performance can be found at:

Scripts we have used to extract performance and plot comparisons can be found in the analysis/ subdirectory.

Further information

If you want further information on these benchmarks, please contact the DiRAC Helpdesk at: [email protected]:1

About

Repository of information and results for the DiRAC Extreme Scaling benchmarks.

Topics

Resources

Stars

Watchers

Forks

Languages