We use cookies to ensure that we give you the best experience on our website. You can change your cookie settings at any time. Otherwise, we'll assume you're OK to continue.

Durham University

Advanced Research Computing

Performance Analysis Methodology Workshop

An ExCALIBUR knowledge exchange activity in collaboration with POP HPC in collaboration with the Computer Science Department and Advanced Research Computing, Durham University, DiRAC and N8 CIR.

This 3-hour virtual workshop aims to give a solid understanding of the basics of performance analysis for research and simulation codes, using high- and low-level metrics.

The workshop can be taken independently, or as follow-up of the Performance Analysis Workshop series.

Workshop will be held on Wednesday (15th of December) 9:00-12:00 UTC on Zoom. The structure and content of the workshop will be as follows:

5 min

Introduction to the POP Centre of Excellence

25 min

Introduction to parallel performance analysis

We describe why it’s often difficult to understand poor parallel performance, introduce the profiling-optimisation cycle, and discuss some challenges associated with tracing-based profiling.

45 min

Introduction to parallel performance analysis using the POP metrics

We’ll introduce high-level parallel efficiency metrics and show examples to illustrate their usefulness in knowing what to look for in trace data.

45 min

MPI parallel performance analysis using the POP metrics

We’ll look at low-level metrics designed specifically for MPI performance analysis that measure performance loss due to:

  • Computation imbalance
  • Time in data transfer
  • MPI dependencies
30 min

POP metrics for OpenMP and hybrid codes

In this session we’ll describe some low-level metrics for:

  • OpenMP
  • Hybrid MPI + OpenMP parallelisation
30 min

Introduction to the BSC tracing Tools & PyPOP

  • Extrae is a relatively lightweight and easy to use tracing tool, the traces generated can be used to calculate the POP metrics.
  • Paraver can be used to view Extrae trace timelines and calculate some simple metrics.
  • Dimemas can simulate performance on an ‘ideal’ interconnect to calculate Transfer Efficiency.
  • PyPOP automates calculation of POP metrics from Extrae traces.


To register, please email with the following information:

  • Name
  • Email
  • Institution and group
  • Job title/role



  • Jon Gibson (NAG)
  • Jonathan Boyle (NAG)
Computer Science Department