Periscope Tuning Framework
This is the official web presence of the Periscope Tuning Framework. The PTF is a suite of tools designed to assist the HPC application developer in the optimization of their application. Below is a quick overview of our main areas of contribution.
The current version, PTF 2.0, is available since January 2016. It is based on the Score-P monitoring infrastructure. Score-P provides performance analysis data and allows to dynamically configure tuning paramters of the application and the execution environment. Score-P replaces PTF’s proprietary monitoring system.
All tools are available free of charge. The best way to get in touch with our tools is to visit us during one of our workshops or contact us directly.
We also maintain a YouTube channel for introduction and video tutorials.
Areas of contribution
Using Periscope, an online instrumentation and measurement tool, the developers can more easily identify the bottlenecks of their application. Periscopes develops hypotheses about the bottlenecks of an application automatically and verifies them during run-time.
Performance analysis with automatically verified hypothesis can give the developer a good starting point and is especially advisable, when the exact bottlenecks are not yet known.
Using Periscope’s various tuning-plugins, it is possible to automatically find the optimal combination of settings such as compiler flags, MPI settings, the number of OpenMP threads in each parallel section, etc.
Auto-tuning is an easy way to gain some performance with minimal effort.
Using Pathway, the whole process of optimizing an application, including for example running jobs on the HPC system, adjusting the job’s settings and recording data can be formalized and partially automated.
Automated workflows reduce the potential for errors and eliminate boring work.