Memory Profiling

OData support
Lazányi János Gyula
Department of Measurement and Information Systems

In these days the number of multiprocessor systems gradually increase. Therefore it is an important question whether the algorithms running on these systems are using the resources efficiently. Many algorithms can be easily modified to run on multiple processors, but some (mostly stream-based video and audio processing) can make this task very difficult. Creating a tool that could solve this problem automatically is a very complex task, but it would greatly change the design and implementation of these algorithms.

The first step of modifying the algorithms is an in depth analysis. In this thesis I propose a method that could be used in this step. The implementation of an analysis tool is also discussed.

The studied literature describes some analysis methods, including the dynamic dataflow graphs. Based on this method I created a new, more usable analysis process, that aggregates the dataflow graphs to extract their essence.

I tested the developed tool by analyzing a JPEG decoder algorithm. During this analysis I found, that the aggregated dataflow graph is a very powerful tool for discovering the internal connections of a given algorithm.

In conclusion the analysis tool described in this thesis is very effective and usable, and might serve as a base for developing the aforementioned automatic tool.


Please sign in to download the files of this thesis.