Predictive Thermal Scheduling for Compute IP Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current operating system schedulers do not consider thermal and energy efficient usage of compute IP blocks when assigning workloads, leading to potential overheating, performance penalties, and the need for thread migration in fan-less systems.
Innovation Solution
Implement a thermally aware and power efficient workload scheduler that predicts temperatures and power consumption of compute IP blocks to optimize workload assignment, reducing the need for thread migration and thermal throttling by selecting IP blocks that maintain lower temperatures and minimize power leakage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Temperature
If current operating system schedulers assign workloads to compute IP blocks without considering temperature, then workload assignment is simple and fast, but compute IP blocks overheat leading to performance penalties and thread migration
Solution Approach 1:
The scheduler predicts future temperatures of compute IP blocks before assigning workloads, allowing the system to proactively avoid overheating by selecting cooler blocks in advance. This preliminary temperature prediction enables the scheduler to make informed decisions that prevent thermal issues before they occur, rather than reacting after overheating happens.
Solution Approach 2:
The scheduler incorporates temperature feedback from compute IP blocks to dynamically adjust workload assignments. By continuously monitoring actual and predicted temperatures, the scheduler can adapt its decisions to maintain optimal thermal conditions, creating a closed-loop control system that balances workload distribution with thermal management.
2Productivity
If workloads are assigned without thermal awareness, then system operation is straightforward, but thermal throttling and thread migration occur causing performance loss
Solution Approach 1:
The scheduler performs preliminary temperature prediction and evaluates multiple candidate compute IP blocks before finalizing workload assignment. This advance planning ensures that workloads are directed to blocks that will remain within safe temperature ranges, preventing thermal throttling and maintaining both high productivity and system reliability.
Solution Approach 2:
The scheduler changes the decision-making parameters by incorporating predicted temperature as a key factor alongside traditional metrics like load balance and CPU affinity. This parameter expansion allows the system to optimize for both performance and thermal stability simultaneously, avoiding the need for thread migration and thermal throttling.
3Loss of energy
If the scheduler selects compute IP blocks without predicting power consumption, then assignment is simple, but power leakage increases reducing energy efficiency
Solution Approach 1:
The scheduler predicts power consumption and temperature of compute IP blocks before workload assignment, allowing proactive optimization of energy efficiency. By evaluating predicted power leakage alongside other factors, the scheduler can select blocks that minimize energy waste while maintaining system performance.
Solution Approach 2:
The scheduler uses power consumption feedback to continuously refine workload assignments and reduce overall system power leakage. By monitoring actual power usage against predictions, the system can adapt its scheduling decisions to minimize energy waste, creating a feedback-driven energy optimization mechanism.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Methods, apparatus, and systems to dynamically schedule a workload to among compute blocks based on temperature are disclosed. An apparatus to schedule a workload to at least one of a plurality of compute blocks based on temperature includes a prediction engine to determine (i) a first predicted temperature of a first compute block of the plurality of compute blocks and (ii) a second predicted temperature of a second compute block of the plurality of compute blocks. The apparatus also includes a selector to select between the first compute block and the second compute block for assignment of the workload. The selection is based on which of the first and second predicted temperatures is lower. The apparatus further includes a workload scheduler to assign the workload to the selected one of the first or second compute blocks.