13 Conclusion

We wanted to show that it is possible to implement a portable and scalable real-time Java extension with hard real-time capabilities.

The platform dependent code is relatively small and manageable. It is much easier to port JaRTS to a new platform than a complex JVM implementing RTSJ.

Currently the TimeSys RTSJ reference implementation is based on the CVM and the Foundation Profile libraries [20]. CVM is too big for resource limited systems (more than 2.5 MByte for CVM with Foundation Profile) where only a very small library would be sufficient.

In addition to this JaRTS can be used with any JVM implementing JNI. So the applications can use all available Java libraries. Compared to this CVM is very limited and, for example, does not provide graphics yet.

The following list summarizes the JaRTS features:

• Can be used on operating systems with Interrupt Abstraction (two kernel solution) easily
• Can be used on small embedded systems not requiring a full Java environment
• Easy to port to different systems
• Only some operating system dependent files used by the JaRTS compiler have to be adapted (rest: ANSI-C, compiler available for almost every operating system)
• An off-the-shelf JVM which is already available for most systems is used for the non-real-time parts. This JVM can support all known Java libraries.

Since JaRTS currently is a prototype implementation there is space for improvement in terms of memory and performance. The bock compiler was chosen to get to a first implementation very quickly. The translation was not tuned to be efficient and resource saving. Nevertheless with the current implementation it is possible to generate control programs with very low latency times and an acceptable performance.