Appendix To Reducing Tardiness Under Global Scheduling by Splitting Jobs

T split i = T base i si , (A.9) and we add a prime to any variable to indicate the version that accounts for overheads. Aside from the modifications described below, we model split tasks as full tasks with worst-case execution time C i and period T split i , so that subjobs function as jobs. As discussed in Sec. 5, several overheads cannot be incurred by all subjobs. The event latency ∆ appears in (8), but can only affect an initial subjob. Because it measures an event that occurs outside the CPU, this particular overhead is actually incurred by the base job rather than the subjob. Therefore, instead of charging ∆ to every subjob period by defining