Dynamic Adaptive Scheduling for Heterogeneous SoC Energy Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional operating system (OS) schedulers for domain-specific systems-on-chip (DSSoCs) have significant runtime overhead, undermining the potential performance and energy efficiency of these specialized processors, as their overhead can be orders of magnitude larger than the execution time of tasks.
Innovation Solution
A dynamic adaptive scheduling (DAS) framework that combines the benefits of a fast, low-overhead scheduler and a sophisticated, high-performance but high-overhead scheduler, dynamically switching between them based on system resources and workload characteristics using a runtime preselection classifier to minimize scheduling overhead and maximize performance and energy efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OS schedulers are used for DSSoCs, then scheduling decisions can be made with comprehensive analysis, but scheduling overhead becomes orders of magnitude larger than task execution time
Solution Approach 1:
The patent implements a dynamic scheduler that adapts its behavior based on system state. It uses runtime performance monitoring to identify when tasks are bottlenecked by scheduling overhead versus execution time, and dynamically switches between aggressive scheduling (when overhead dominates) and conservative scheduling (when execution time dominates), optimizing the balance between decision quality and overhead
Solution Approach 2:
The scheduling system is segmented into multiple components: a fast lightweight scheduler for routine decisions, a performance monitoring module that tracks execution metrics, and a dynamic adjustment mechanism. This segmentation allows the system to use simple scheduling most of the time while occasionally invoking more sophisticated scheduling when performance data indicates it would be beneficial
2Loss of time
If fast low-overhead schedulers are used, then scheduling overhead is minimized, but scheduling decision quality deteriorates
Solution Approach 1:
The patent implements feedback loops that continuously monitor task execution performance, scheduling overhead, and system state. This feedback information is used to dynamically adjust scheduling parameters and switch between scheduling strategies, ensuring that the scheduler adapts to actual system conditions rather than using fixed scheduling policies
3Measurement precision
If sophisticated high-performance schedulers are used, then scheduling decision quality improves, but energy consumption increases
Solution Approach 1:
The scheduler dynamically adjusts its sophistication level based on monitored performance metrics. When execution time is significantly longer than scheduling overhead, the system switches to simpler, lower-energy scheduling strategies. When scheduling overhead becomes comparable to or exceeds execution time, the system activates more sophisticated scheduling algorithms, thereby optimizing energy consumption relative to actual performance needs
Data Source
AI summary
Provided herein are dynamic adaptive scheduling (DAS) systems. In some embodiments, the DAS systems include a first scheduler, a second scheduler that is slower than the first scheduler, and a runtime preselection classifier that is operably connected to the first scheduler and the second scheduler, which runtime preselection classifier is configured to effect selective use of the first scheduler or the second scheduler to perform a given scheduling task. Related systems, computer readable media, and additional methods are also provided.