Examples of Analysis

This page collects some examples of studies carried out with the Paraver tool.

  • Analysis of scalability problems
    A simple example of the first steps that can be used to analyze the poor scalability of a program (comparing metrics like number of instructions, instruction per time unit or memory misses).
     
  • Performance analysis of the Comunity Atmosphere Model (CAM) with Paraver
    This study shows what kind of information and the level of details that can be obtained analyzing a complex program just looking at the paraver tracefile without having a look at the source code. The tool allowed us to extract information on the program characterization (detect inbalances between processes and between iterations, meassure the number of instructions, FMA/FPU ratio...) and the performance achieved (MFLOPS, memory access ratio...).
     
  • POP performance prediction
    Detailed analysis with Paraver This example shows how interoperation between Paraver and Dimemas is an extremely powerful tool to understand the behavior of MPI applications both at the macroscopic and microscopic level. The example shows what we think is the most adecuate way to jointly use Paraver and Dimemas.
     
  • Analyzing the influence of interrupts and system daemons on fine grain applications
    The AIX trace facility allows to collect very low level information on the processes scheduling, system calls... We have developed a translation from AIX trace format to Paraver so now we are able to use all the functionalities of Paraver to analyze the AIX trace data. The example describes the analysis carried out to meassure the interference of the OS on very fine grain applications.
  • Comparing MLP and MPI
    An analysis of two versions of a code that should have behaved similarly and didn't. The study is able to point to the mapping of processes and memory pages to nodes as the cause of the difference. We show how indirect metrics not captured by the tracing package (such as actual L2 miss cost) can be inferred.

Additional information on the methodology of analysis can be found on the MPI+OpenMP Performance Analysis Tips manual available in html format.