Avoiding deadline decay under transient overloads

Deadline scheduling policies based on stochastic computation times perform well in average but fail in presence of transient overloads, causing already scheduled tasks to miss their deadline. This paper presents an approach using medium-term scheduling policies to cope with the problem of transient overload in dynamic real-time systems. The proposed approach has been implemented in a Real-Time Distributed Operating System (RTDOS) running on a Transputer network.<<ETX>>