Task Scheduling for Multi-core and Parallel Architectures by Quan Chen & Minyi Guo

Author:Quan Chen & Minyi Guo
Language: eng
Format: epub
Publisher: Springer Singapore, Singapore

5.3.3.2 Bias Scheduling Algorithm

Bias scheduling algorithm is built on top of the existing scheduler of operating system. It does not change the existing scheduler of operating system in ways that would dictate when to run a thread or change any system properties that the scheduler is trying to maintain such as fairness, responsiveness or real time constraints. It works by preferably scheduling applications on the core type that best suits its bias.

In order to implement bias scheduling, the operating system is modified to support fast-core first scheduling and asymmetry-aware load balancing. The fast-core first scheduling enables maximum performance by scheduling first to idle big cores. The asymmetry-aware load balancing schedules work in big cores proportional to an estimated ratio of big/small performance.

It is worth noting that bias scheduling can be performed on top of any existing scheduling algorithm. Its design tries to minimize changes to the existing scheduler by focusing on two areas: imbalanced system and balanced system as described below.

On an imbalanced asymmetric system, the scheduler tries to migrate a thread from the busiest core to the idlest core. Bias scheduling does not change the way in which these cores are selected. Once they have been identified, the required load that has the highest bias is migrated to the destination core. If the cores have the same type, bias is irrelevant. In some instances the scheduler cannot find a thread with an appropriate bias, and defaults to migrating any thread just as it does without bias.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.