Abstract

Real-time software is commonly written in low level languages such as C and assembler. Such software is often hard to test, maintain, and port to other platforms. To improve this situation, high level languages are introduced in the real-time domain. The Java programming language has many properties that are appealing to real-time system development, e.g. robustness and portability. However, it is difficult to estimate the worstcase execution time of Java code using current implementations. This is due to several properties of Java, e.g. dynamic binding of method calls and automatic memory management. This thesis discusses the problems of introducing automatic memory management in hard real-time environments and proposes a technique called RT-Reference Counting to solve the problems. This technique is being integrated in a native Java compiler called RT-Java, being implemented at the Programming Environments Laboratory (PELAB) at Linköping University. An automatic memory manager for hard real-time systems must be able to guarantee that memory is always available to the application and that all memory management operations are fast and execution time predictable. Software based automatic memory managers for real-time systems exist, but they have problems, e.g. always guaranteeing short deadlines and/or handling large objects. RT-Reference Counting aims at solving these problems. The technique is not yet fully developed, but results so far are promising.

Citations