README for mpi_sync_clocks

This document describes mpi_sync_clocks installation and configuration.

for any question, send an email to: Alexa.nosp@m.ndre.nosp@m..Deni.nosp@m.s@in.nosp@m.ria.f.nosp@m.r.

mpi_sync_clocks is an implementation of synchronized clocks using MPI. It offers:

• interpolated clocks to get a consistent clock accross nodes through interpolation, i.e. computed at the end of the execution. This is the most precise method.
• extrapolated clocks to compute consistent clocks accross nodes in realtime. This method is less precise than interpolation.
• synchronized barrier: a barrier that guarantees to exit on all nodes at the same time, contrary to MPI_Barrier. This is especially useful to benchmark collective operations.

Requirements

• autoconf (v 2.50 or later)
• any MPI library

Installation

• Install follows usual autotools procedure:

  ./autogen.sh
  ./configure [your options here]
  make install
  

• no compilation is needed. The library consists only of headers.

Documentation

• To locally generate doxygen documentation:

  % make docs
  

• It is also available online at http://pm2.gitlabpages.inria.fr/pm2/mpi_sync_clocks/doc/.
• Reference API documentation: mpi_sync_clocks public API.