Optimal Task Scheduling Benefits From a Duplicate-Free State-Space (1901.06899v1)

Abstract: The NP-hard problem of task scheduling with communication delays (P|prec,c_{ij}|C_{\mathrm{max}}) is often tackled using approximate methods, but guarantees on the quality of these heuristic solutions are hard to come by. Optimal schedules are therefore invaluable for properly evaluating these heuristics, as well as being very useful for applications in time critical systems. Optimal solving using branch-and-bound algorithms like A* has been shown to be promising in the past, with a state-space model we refer to as exhaustive list scheduling (ELS). The obvious weakness of this model is that it leads to the production of large numbers of duplicate states during a search, requiring special techniques to mitigate this which cost additional time and memory. In this paper we define a new state-space model (AO) in which we divide the problem into two distinct sub-problems: first we decide the allocations of all tasks to processors, and then we order the tasks on their allocated processors in order to produce a complete schedule. This two-phase state-space model offers no potential for the production of duplicates. We also describe how the pruning techniques and optimisations developed for the ELS model were adapted or made obsolete by the AO model. An experimental evaluation shows that the use of this new state-space model leads to a significant increase in the number of task graphs able to be scheduled within a feasible time-frame, particularly for task graphs with a high communication-to-computation ratio. Finally, some advanced lower bound heuristics are proposed for the AO model, and evaluation demonstrates that significant gains can be achieved from the consideration of necessary idle time.