Deterministic Execution of Java’s Primitive Bytecode Operations
Fridtjof Siebert IPD, Universität Karlsruhe
Am Fasanengarten 5 76128 Karlsruhe, Germany

Andy WalterForschungszentrum Informatik (FZI)
Haid- und Neu Straße 10-14 76131 Karlsruhe, Germany

siebert@ira.uka.de

anwalt@ira.uka.de

This work was partially funded by the DFG program GRK 209--------

ABSTRACT

For the application of Java in realtime and safety critical domains, an analysis of the worst-case execution times of primitive Java operations is necessary. All primitive operations must either execute in constant time or have a reasonable upper bound for their execution time. The difficulties that arise for a Java virtual machine and a Java compiler in this context will be presented here. This includes the implementation of Java’s class and interface model, class initialization, monitors and automatic memory management. A new Java virtual machine and compiler that solves these difficulties has been implemented and its performance has been analysed.

Although incremental garbage collection techniques can help to reduce the likelihood for a blocking garbage collection pause, they can not guarantee it. It can still occur that the collector does not make sufficient progress and does not catch up with the application. Consequently, the system can fail or require long blocking pauses to recycle memory or defragment the heap. A deterministic implementation of Java must provide means to determine worst-case execution times for Java’s primitive operations. The dynamic structure of Java, with inheritance, virtual method calls and multiple-inheritance for interfaces, poses several difficulties for the implementation. The time required for calls or type checks must be limited and statically determinable. Further difficulties are caused by the dynamic nature of Java’s class loading and initialization