ABSTRACT
Abstract In this chapter, we propose a general framework for deriving efficient (polynomial-time) algorithms for steady-state scheduling. In the context of large scale platforms (grids or volunteer computing platforms), we show that the characteristics of the resources (volatility, heterogeneity) limit their use to large regular applications. Therefore, steady-state scheduling, that consists in optimizing the number of tasks that can be processed per time unit when the number of tasks becomes arbitrarily large, is a reasonable setting. In this chapter, we concentrate on bag-of-tasks applications, and on scheduling collective communications (broadcast and multicast). We prove that efficient schedules can be derived in the context of steady-state scheduling, under realistic communication models that take into account both the heterogeneity of the resources and the contentions in the communication network.
