NUMA-aware scheduling for both memory- and compute-bound tasks

There has been much work in NUMA-aware (Non-Uniform Memory Access) scheduling the past decade, all authors presenting different schedulers and considers the major problems with today schedulers, which are mostly aimed at UMA (Unified Memory Access) machines or not optimized for NUMA machines. This paper is aimed at summarizing the available literature about NUMA-aware scheduling and extract guidelines for how to schedule compute- and memory-bound tasks in a NUMA-aware fashion. This can be done by using different techniques to distribute the data among the available nodes and by fully utilizing all of the memory controllers sockets. From the discussion and theory, it is possible to form eight guidelines that can be used to write a NUMA-aware scheduler, formed from the main ideas of the available literature