Investigating the effect of processor affinity on uniform parallel machine scheduling

In this paper we study the effect of processor affinity on real-time scheduling algorithms in uniform multiprocessor environments. For this reason we compare our two previously proposed algorithms called EFDF and FLLF, with two well know algorithms, namely global EDF and global LLF. Despite its non-optimality, EDF is an appropriate algorithm to use on multiprocessors. However, the performance of the traditional algorithms such as EDF and LLF significantly degrades in overloaded situations. Moreover, they produce a relatively large number of migrations which may prove unacceptable for use on some parallel machines. It is shown that our proposed algorithms not only demonstrate a performance as good as that of EDF in non-overloaded conditions but also overcome its deficiencies in overloaded situations in many aspects. Furthermore, they impose much less overhead on the system.