The studied problem is the jointly scheduling of hard periodic tasks with soft aperiodic events, where the response times of soft tasks have to be as low as possible while the warranty to meet their deadlines has to be given to hard tasks. A lot of theoretical solutions have been proposed these past two decades but we are interested on the implementability of these solutions under the real-time specification for Java (RTSJ), without changing the scheduler. This led us to adapt the existing algorithms to operate at a user land level in the system, to propose some optimizations and counter measures in order to balance the lost of performances and finally to set up an approximate slack stealer algorithm specifically designed to take into account RTSJ restrictions. We propose new classes to extend the RTSJ API's to implement these mechanisms and some minor modification suggestions to existing ones as a feed back from our RTSJ experiences. We demonstrates the efficiency of the modified algorithms through extensive simulations and the implementability on available RTSJ compliant virtual machine by an overhead measure in real situation with the RTSJ JamaïcaVM from Aïcas. We also measure the overhead on LejosRT, an RTSJ compliant firmware for Lego Mindstorms NXT in development.
展开▼
