AI Resource Scheduling via Autonomous Task Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In AI resource scheduling, existing methods fail to effectively utilize AI resources across devices due to performance differences and varying resource requirements, leading to idle or underutilized devices, especially in large-scale or multi-scenario deployment environments.
Innovation Solution
An AI resource scheduling method that involves obtaining scheduling information from multiple AI devices, determining busy and idle devices based on task and resource information, and transferring tasks from busy devices to idle ones with sufficient resources, optimizing resource allocation and utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If AI tasks are independently executed by each device with fixed assignments, then device simplicity and ease of operation are maintained, but AI resource utilization efficiency deteriorates due to idle devices and uneven workload distribution
Solution Approach 1:
Each AI device autonomously determines its own workload status by comparing its current task queue length against a threshold value, and independently initiates task transfers to peer devices without requiring centralized scheduling control. This self-service mechanism resolves the contradiction by enabling resource utilization improvement through distributed autonomous decision-making rather than complex centralized coordination.
Solution Approach 2:
The patent merges the scheduling function into the existing task execution framework by having devices share task queue information and autonomously coordinate transfers. This combining of scheduling and execution functions allows resource optimization without adding separate complex scheduling infrastructure, thus improving productivity while avoiding excessive system complexity.
2Measurement precision
If high-configuration devices are assigned more AI tasks, then task execution precision is improved, but resource balance deteriorates causing idle low-configuration devices
Solution Approach 1:
The patent implements dynamic task allocation where devices continuously monitor their task queue lengths and autonomously transfer tasks to peer devices when thresholds are exceeded. This dynamic adjustment mechanism allows high-configuration devices to maintain precision for complex tasks while simultaneously balancing the workload to prevent idle low-configuration devices, thus resolving the contradiction between precision and overall resource utilization.
Solution Approach 2:
The system changes the parameter of task distribution by introducing threshold-based autonomous transfer mechanisms and workload balance considerations. Instead of static assignment, devices adjust task allocation based on real-time parameters such as queue length and configuration differences, enabling both high precision execution and improved overall resource utilization.
3Productivity
If low-configuration devices execute AI tasks, then resource distribution is improved, but execution precision deteriorates
Solution Approach 1:
The dynamic threshold mechanism allows low-configuration devices to execute tasks when it benefits overall system balance, while high-configuration devices automatically accept transfers of computationally intensive tasks. This dynamic adaptation resolves the contradiction by enabling low-configuration devices to contribute to resource utilization without permanently sacrificing precision for tasks that require higher computational power.
Solution Approach 2:
The patent introduces an intermediary mechanism where devices act as mediators for task transfers. When a low-configuration device has an overloaded queue, it can transfer tasks to peer devices with available capacity, ensuring that tasks are executed by appropriately configured devices while maintaining overall resource distribution balance.
4Productivity
If devices autonomously transfer tasks to peer devices, then resource utilization is improved, but system coordination complexity increases
Solution Approach 1:
Each device autonomously determines when task transfers are needed by comparing its queue length against thresholds and independently initiates transfers to peer devices. This self-service approach improves resource utilization while avoiding complex centralized coordination, as each device manages its own workload without requiring sophisticated system-wide coordination mechanisms.
Solution Approach 2:
The system employs feedback mechanisms where devices share task queue information with peers, and transfer decisions are based on feedback from queue length comparisons. This feedback-driven autonomous coordination enables improved resource utilization through simple threshold-based decisions rather than complex coordination protocols.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
This application provides an artificial intelligence resource scheduling method and apparatus, a storage medium, and a chip. The scheduling method includes: obtaining scheduling information of each artificial intelligence device in a plurality of AI devices, where the scheduling information includes at least one of the following: task information of one or more AI tasks executed by each AI device or resource information of each AI device; determining a first device and a second device from the plurality of AI devices based on the scheduling information of each AI device; and transferring at least one AI task in one or more AI tasks of the first device to the second device. In this solution, the plurality of AI devices can collaborate by scheduling AI resources, thereby improving utilization of the artificial intelligence resources.