| Summary: | In this thesis we addressed the question of whether taint tracking could be used to help developers make better, faster decisions when improving existing web applications. This is in the context of migration scenarios, where one must respond to increasing demands on an application by optimizing and generally rearchitecting. We wanted to determine if detailed dynamic dataflow traces from web applications could support automated analyses, the results of which would help one to better understand applications and guide one in optimizing them.
To investigate this problem, we identified a set of useful analyses from a search of the literature and from our own experience with web applications. These analyses were developed to run automatically over taint tracking data, producing output which should be immediately useful to non-expert users.
Two real applications were chosen for analysis in order to determine two important things. First, that we could write our analyses to automatically identify their targets and produce comprehensible results. Second, that the targets actually existed in real applications.
In the end our analyses were successful, in many cases producing clean results which concisely described non-trivial properties of the applications and possible optimizations to them. By focusing on how data moves through a system, we found a natural fit for understanding its workings. The biggest difficulties manifested as a need for further automation, to take complicated analysis results and simplify them. Even with many challenges, we believe that our techniques are valuable for helping developers, and should be more thoroughly studied.
|
|---|
