On the Dominance of Minimum-Parallelism Multiprocessor Supply
Kecheng Yang, James H. Anderson · 2016
Many approaches have been proposed to enable disparate real-time software components to share a physical multiprocessor platform by giving each component the "illusion" of executing on a dedicated virtual platform. Such an illusion is supported by specifying a supply interface that indicates how computation time is made available to a component over time. A number of approaches for defining such interfaces have been proposed: so many that sifting through them all can be confusing for the practitioner. In the case of soft real-time applications, one particular proposed interface-minimum-parallelism (MP) supply-has been shown to enable the co-scheduling of different components with no utilization loss. In the case of hard real-time applications, it follows from prior work that MP supply easily dominates other choices if the simplifying assumption is made that supply is allocated on different processors using a common, synchronized allocation period. The main contribution of this paper is to show that the dominance of MP supply is retained if this simplifying assumption is removed, provided the period of allocation is defined properly. This result suggests that MP supply should be the focus in future work on real-time multiprocessor virtualization.