Introduction to Performance Engineering

Content

HPC systems are expensive and increasingly widely used. Many applications, whether under development or already in use, can be installed on these systems. In order to use the valuable resources offered by these systems, care must be taken, not only in development of efficient algorithms, but also in many hardware, programming and deploying aspects. An attempt will be made to discuss all important aspects, and those which are mainly relevant for HPC systems, such as load balancing and communication overhead, will be focussed upon in more detail

Overview of factors which influence performance:

• Basics of parallel computer architecture and topology
• Parallelization: current methods, Amdahl’s Law
• Performance bottlenecks
• Methods and tools for analysis
• Examples (LIKWID, Vampir & Scalasca)

Possible tools:

• LIKWID
• Score-P
• Scalasca
• Vampir

Learning goal

• The course is intended to give a general view of the most relevant aspects for efficient use of resources by applications on HPC systems.
• The participants will learn basic terminology and will be referred to other sources for further indepth information and training.

Skills

Trainer

• Dr. Jack Ogaja

Next appointment