Homec4science
Diffusion lammps (lammm-master-deprecated) Browse

lammps/doc/doc2/timers.html
eee480c020ealammm-master-deprecated

Last Change
Raw File

timers.html
View Options

<HTML>
<CENTER><A HREF = "http://lammps.sandia.gov">LAMMPS WWW Site</A> - <A HREF = "Manual.html">LAMMPS Documentation</A> - <A HREF = "Section_commands.html#comm">LAMMPS Commands</A>
</CENTER>
<HR>
<H3>timestep command
</H3>
<P><B>Syntax:</B>
</P>
<PRE>timers args
</PRE>
<LI><I>args</I> = one or more of <I>off</I> or <I>loop</I> or <I>normal</I> or <I>full</I> or <I>sync</I> or <I>nosync</I>
<PRE> <I>off</I> = do not collect and print timing information
<I>loop</I> = collect only the total time for the simulation loop
<I>normal</I> = collect timer information broken down in sections (default)
<I>full</I> = like <I>normal</I> but also include CPU and thread utilzation
<I>sync</I> = explicitly synchronize MPI tasks between sections
<I>nosync</I> = do not synchronize MPI tasks when collecting timer info (default)
</PRE>
<P><B>Examples:</B>
</P>
<PRE>timers full sync
timers loop
</PRE>
<P><B>Description:</B>
</P>
<P>Select to which level of detail LAMMPS is performing internal profiling.
</P>
<P>During regular runs LAMMPS will collect information about how much time is
spent in different sections of the code and thus can provide valuable
information for determining performance and load imbalance problems. This
can be done at different levels of detail and accuracy. For more
information about the timing output, please have a look at the <A HREF = "Section_start.html#start_8">discussion
of screen output</A>.
</P>
<P>The <I>off</I> setting will turn all time measurements off. The <I>loop</I> setting
will only measure the total time of run loop and not collect any detailed
per section information. With the <I>normal</I> setting, timing information for
individual sections of the code are collected and also information about
load imbalances inside those sections presented. The <I>full</I> setting adds
information about CPU utilization and thread utilization, when multi-threading
is enabled.
</P>
<P>With the <I>sync</I> setting, all MPI tasks are synchronized at each timer call
and thus allowing to study load imbalance more accuractly, but this usually
has some performance impact. Using the <I>nosync</I> setting this can be turned
off (which is the default).
</P>
<P>Multiple keywords can be provided and for keywords that are mutually
exclusive, the last one in that group is taking effect.
</P>
<P>IMPORTANT NOTE: Using the <I>full</I> and <I>sync</I> options provides the most
detailed and accurate timing information, but also can have a significant
negative performance impact due to the overhead of the many required system
calls. It is thus recommended to use these settings only when making tests
to identify the performance. For calculations with few atoms or a very
large number of performance, even using the <I>normal</I> setting can have
a measurable performance impact. It is recommended in those cases to use
the <I>loop</I> or <I>off</I> setting.
</P>
<P><B>Restrictions:</B> none
</P>
<P><B>Related commands:</B>
<A HREF = "run.html">run post no</A>, <A HREF = "kspace_modify.html">kspace_modify fftbench</A>
</P>
<P><B>Default:</B>
</P>
<P>timers normal nosync
</P>
</HTML>
c4science · Help