GstShark profiling: a real-life example

GstShark is a profiling and benchmarking tool for GStreamer pipelines. GstShark is an ongoing OpenSource project by RidgeRun which serves as a front-end for the GstTrace subsystem. GstShark presents raw traces as higher level data such as scheduling and processing time, bitrate, framerate, CPU usage and much more. This data is saved in a standard low-footprint format designed for efficient tracing. The captured data can be plotted and visualized using the tools included in the project, as well as third party tools. GstShark is the result of years of experience tuning and optimizing GStreamer pipelines in resource-limited systems and is key tool RidgeRun engineers use to dispel the myth that GStreamer is slower that an inflexible custom created streaming media application.

In this session GstShark will be used to optimize a low-performance WebRTC streaming pipeline in an NVidia Tegra embedded platform. It will be shown how the different measurements can be used to identify processing bottlenecks, sources of latency and general scheduling problems. By using comprehensive data plots, the pipeline internals are exposed, revealing information before hidden and allowing you to tune pipelines in a more informed, deterministic way.

Michael Grüner is the Tech Leader at RidgeRun, a GNU/Linux based embedded software development company. GStreamer and multimedia have been his main areas of focus. Michael has a masters degree in Digital Signal Processing and, among other interests, likes OpenGL, CUDA, OpenCL. Michael is always looking for ways to implement efficient, real-time DSP algorithms using GStreamer on embedded platforms.