Task Graph Neural Network Scheduling for Multi-Processor Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multi-processor embedded systems, existing methods for assigning tasks to processors to ensure real-time processing, such as bin packing problems, often fail to provide optimal solutions due to the complexity of determining all possible assignment cases, leading to suboptimal task scheduling.
Innovation Solution
An electronic device determines features of tasks based on their period and execution time, generates a graph representing task relationships, and uses a trained neural network model to assign tasks to processors, ensuring tasks are completed within deadlines by considering priority interference and constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional bin packing methods are used to assign tasks to processors, then the assignment process is simplified, but optimal scheduling solutions cannot be guaranteed due to the complexity of determining all possible assignment cases
Solution Approach 1:
The patent replaces traditional mechanical bin-packing algorithms with a neural network-based intelligent system. The neural network learns optimal task-to-processor assignments by training on task graphs that represent temporal and resource constraints, substituting deterministic but suboptimal mechanical algorithms with adaptive intelligent computation that achieves superior scheduling optimality without proportionally increasing system complexity
Solution Approach 2:
The patent transforms the task assignment problem by changing parameters from simple task characteristics to comprehensive task graphs that include temporal constraints, resource requirements, and inter-task dependencies. This parameter transformation enables the neural network to capture complex relationships and determine optimal assignments that traditional methods cannot achieve
2Manufacturing precision
If all possible task assignment cases are determined to ensure optimality, then optimal scheduling is achieved, but the computational complexity and time required increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network on comprehensive task graphs and assignment scenarios before actual scheduling. This pre-training phase allows the system to learn optimal assignment patterns and constraints in advance, so that during runtime, the system can quickly determine optimal assignments without exhaustively evaluating all possible cases, thus reducing computational complexity while maintaining optimality
Solution Approach 2:
The patent uses copying by creating task graphs that replicate the structure and constraints of actual task sets. These graphical representations serve as simplified models that the neural network can process efficiently, allowing the system to determine optimal assignments without directly analyzing the full complexity of all possible assignment cases
3Reliability
If task assignment methods focus on ensuring real-time properties, then deadline compliance is improved, but the ability to achieve optimal resource utilization is reduced
Solution Approach 1:
The patent merges previously separate concerns of real-time deadline compliance and resource utilization optimization into a unified neural network-based assignment system. By simultaneously considering both constraints in the task graph representation and training process, the system produces assignments that satisfy real-time requirements while achieving optimal resource utilization, eliminating the trade-off between reliability and productivity
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An electronic device includes: one or more first processors configured to: determine features of a plurality of nodes corresponding to a plurality of tasks comprised in a task set, based on a period and an execution time of each of the plurality of tasks; determine one or more edges between the plurality of nodes corresponding to a relationship between the plurality of tasks; and generate a graph corresponding to the task set based on the features of the plurality of nodes and the one or more edges between the plurality of nodes, wherein the plurality of tasks is executed according to a deadline which is set for each of the plurality of tasks in one or more second processors to which the plurality of tasks is assigned.