Thermally-Aware Process Scheduling for Thermal Hotspot Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional power management algorithms in processing devices struggle to maintain operating temperatures within a safe range, especially when handling high-power workloads, leading to thermal throttling and potential damage due to uneven thermal density and sensitivity across the device.
Innovation Solution
Implementing thermally-aware scheduling techniques that introduce resource contention between process threads on shared compute units to reduce thermal density and sensitivity, allowing for increased operating frequency and improved performance without exceeding maximum temperatures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional power management algorithms increase operating frequency to improve performance, then productivity increases, but temperature rises causing thermal throttling and reliability issues
Solution Approach 1:
The patent applies local quality by introducing thermal sensitivity weights that vary across different compute units based on their local thermal characteristics. Each compute unit is assigned a thermal sensitivity value that reflects its specific thermal environment, allowing the scheduler to make localized scheduling decisions that account for spatial variations in thermal density and sensitivity across the SOC substrate.
Solution Approach 2:
The patent implements dynamics by continuously updating thermal sensitivity weights and adjusting scheduling decisions in real-time based on changing thermal conditions. The thermal sensitivity weights are dynamically recalculated based on current thermal density measurements and historical thermal data, enabling the system to adapt to varying workload patterns and thermal environments during operation.
2Productivity
If workloads are distributed across multiple compute units to improve performance, then productivity increases, but thermal density increases causing thermal hotspots
Solution Approach 1:
The patent implements feedback by continuously monitoring thermal density measurements from temperature sensors and using this information to adjust scheduling decisions. The system measures thermal density at various locations, calculates thermal sensitivity weights based on these measurements, and feeds this information back into the scheduling algorithm to dynamically redistribute workloads away from thermal hotspots and toward cooler regions.
Solution Approach 2:
The patent applies asymmetry by introducing asymmetric thermal sensitivity weights that reflect the non-uniform thermal characteristics of different compute unit locations. Rather than treating all compute units equally, the system assigns different weights based on their specific thermal environments, with compute units in hotter or more thermally sensitive regions receiving different scheduling priorities compared to those in cooler regions.
3Temperature
If thermal sensitivity weights are introduced to improve thermal management, then temperature control improves, but device complexity increases
Solution Approach 1:
The patent implements parameter changes by introducing thermal sensitivity weights as additional parameters in the scheduling decision process. These weights are calculated based on thermal density measurements and historical thermal data, and they modify the traditional scheduling parameters to account for thermal considerations. The approach transforms the scheduling problem from a simple performance-optimization task to a multi-parameter optimization that includes thermal constraints.
Data Source
AI summary
A scheduler is presented that can adjust, responsive to a thermal condition at the processing device, a scheduling of process threads for compute units of the processing device so as to increase resource contentions between the process threads.