Adaptive partitioning of real-time tasks on multiple cores

What is it about?

When scheduling processes or threads on multiple CPU cores, the strategy used to assign tasks to cores can heavily affect the application's performance. This is particularly critical when the application is characterized by temporal constraints: if the application's tasks are executed on the "wrong" core, the constraints are not respected. This paper presents a new strategy (named "Adaptive Partitioning") for dynamically associating tasks to cores. The new adaptive partitioning strategy tries to get the good properties of the two most commonly used strategies (global scheduling and partitioned scheduling) while avoiding their drawbacks.

Featured Image

Why is it important?

The migration strategy used to schedule real-time tasks on multiple cores has a huge impact on performance, and today there still are multiple open issues. We believe this paper shows a practical way to address such issues, that can be easily implemented in a real OS kernel and provide good performance.