Horst F. Wedde, D. C. Daniels und D. M. Huizinga
Efficient Distributed Resource Scheduling for Adaptive Real-Time Operating System Support
Advances in Computing and Information - ICCI’91, Hrsg.: F. Dehne, Nr. 497, Springer-Verlag, 1991
In this paper we are concerned with real-time operating system services which provide for deadlock-free and starvation-free access to distributed resources that may be replicated or exist in different versions. We define and discuss novel distributed resource scheduling algorithms which provide the support for efficient real-time operations on replicated distributed files, realizing them in the manner of distributed atomic real-time transactions. After constructing a base line algorithm, called Simultaneous Enqueuing, we state formally its relevant properties: starvation-freeness and a minimal message overhead. We then enhance this algorithm by mechanisms which allow for promoting tasks ahead of resource competitors, e.g. because of an earlier deadline. This leads to the Hurry-Up version of Simultaneous Enqueuing. We performed a first series of simulation experiments under varying resource contention and with varying upper bounds for the task laxity, using Partial Ordering resource scheduling as a benchmark. The results are very favorable for the new algorithms which are implemented as part of the kernel services of the real-time version of the distributed operating system DRAGON SLAYER, designed to support its distributed and adaptive replicated file system MELODY.