Multi-Processor Task Scheduling Using a Contention Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-core processors (MCPs) face challenges in safety-critical systems due to contention on shared resources, which complicates task execution and certification, especially in avionics, as existing scheduling methods struggle to predict and mitigate contention delays effectively.
Innovation Solution
A method involving a contention model is used to dynamically query and refine task schedules, accounting for contention delays by generating revised schedules that minimize make-span and improve responsiveness, utilizing a task contention model (TCM) to estimate worst-case execution times and optimize task mapping and scheduling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If tasks are scheduled on multi-processor systems to improve throughput, then productivity increases, but contention delays on shared resources cause execution time variability and reduce reliability
Solution Approach 1:
The contention model is built beforehand to predict resource contention delays. The scheduler queries this pre-built model during task scheduling to obtain accurate execution time estimates, allowing deterministic scheduling decisions to be made in advance while still enabling parallel task execution on multiple processors.
Solution Approach 2:
The system uses feedback from the contention model about predicted delays to adjust task scheduling decisions. The scheduler iteratively refines task assignments based on contention delay predictions, ensuring that the final schedule accounts for resource contention while maintaining high processor utilization and throughput.
2Ease of operation
If traditional scheduling methods are used to simplify scheduling processes, then ease of operation is maintained, but they fail to predict and mitigate contention delays effectively, reducing productivity
Solution Approach 1:
The contention model acts as an intermediary between the simple scheduling interface and the complex multi-processor system. It automatically predicts resource contention delays and provides this information to the scheduler, which then makes informed decisions without requiring complex manual analysis or system configuration.
Solution Approach 2:
The system automatically builds and queries the contention model without requiring manual intervention. The scheduler self-adjusts task assignments based on predicted contention delays, enabling efficient and deterministic scheduling while maintaining ease of operation through automated decision-making.
3Measurement precision
If detailed contention analysis is performed to improve execution time prediction accuracy, then measurement precision increases, but the complexity of the scheduling system increases
Solution Approach 1:
The contention model is built in advance by analyzing the multi-processor system architecture and resource sharing patterns. This pre-computation captures detailed contention characteristics without requiring complex real-time analysis, enabling accurate execution time predictions while keeping the runtime scheduler relatively simple.
Solution Approach 2:
Instead of directly analyzing the complex multi-processor system in real-time, the system creates a simplified contention model that copies and represents the essential resource contention characteristics. This model can be queried efficiently to obtain accurate delay predictions without the computational complexity of full system analysis.
Data Source
AI summary
A computer-implemented method schedules a plurality of tasks for execution by a multi-processor system. A first schedule is generated that assigns each task of the plurality of tasks a time window and a processor of the multi-processor system. A contention model for the multi-processor system is queried to determine a contention delay for the assignment of tasks to processors according to the first schedule. The contention delay determined from the contention model is used to generate, from the first schedule, a revised schedule that assigns each task a time window and a processor of the multi-processor system, wherein the revised schedule is determined in dependence on the determined contention delay.