Queueing analysis in dynamic distributed real-time systems

The effect of queueing on a dynamic distributed real-time system, in which jobs have hard deadline requirements, is studied. A job can be guaranteed to be completed before its deadline expires; it is rejected otherwise. An analytic model is presented to calculate the probability that an arbitrary arriving job will be guaranteed. An embedded Markov chain is first established, followed by a probability analysis. In the model, only one copy of the resource is considered. If more than one copy exists in the distributed real-time system, the scheduling discipline for incoming jobs among these copies will affect the probability that a job can be guaranteed. It is shown that a round-robin scheduling, in which arriving jobs are assigned to all the copies of the resource in a cyclic order, is the best strategy. In this strategy, a job that cannot be guaranteed by its assigned copy can never be guaranteed by other copies either. However, it requires a centralized controller to assign the load. The performance issue for cases with multiple copies of a resource is noted with respect to various strategies. The model is nevertheless useful for further research where multiple copies will be taken into account or for the design of a real-time system where jobs with a hard deadline requirement arrive to the system dynamically and nonperiodically