Garbage collection in the reactive deadline-driven environment of Timber

Software development for small, real-time and resource constrained, embedded systems is becoming increasingly complex. Multi-threading and object- orientation are examples of complicating factors. Even though automated memory management has shown significant advantages, it is commonly considered to be too costly and unpredictable for real-time systems. Timber is a programming language that is based upon a reactive, deadline- driven, concurrent, and object-oriented programming model. It is furthermore based on a functional programming paradigm. In this thesis we put forward a generic garbage collector for the run-time system of Timber. It is based upon a copying collector scheme and it especially makes use of the distinction between mutable and immutable data (induced by the language semantics), as well as the controlled accessibility of mutable data. Our collector is transparent, which means that it never preempts the application, and furthermore is incremental with extremely fine granularity. Minimal synchronization is needed between the application and the collector. Allocation takes an extremely small and constant amount of time (incrementing a pointer). Due to the reactive environment along with deadline-driven scheduling it will never compete for execution time with the application. Sufficient collector progress must thus be guaranteed by static analysis, and we address issues concerning this by briefly discussing memory usage behavior attributes (as a starting point for future work).Validerat; 20101217 (root