Edge Task Dispatching Using Online Learning and Deep Reinforcement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge computing systems face challenges in dynamically dispatching tasks to optimal edge servers without causing network congestion or server overload, and in scheduling tasks to minimize average response time while maintaining efficiency and fairness, due to dynamic network conditions and limited resources.
Innovation Solution
An online task dispatching and scheduling method (OTDS) that combines online learning (OL) and deep reinforcement learning (DRL) techniques, using a multi-armed bandit framework for task dispatching and a weighted, multi-queue Round Robin method with DQN for scheduling, to dynamically allocate resources based on real-time network conditions and task time-sensitivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If tasks are offloaded to cloud data centers, then computation-intensive tasks can be accommodated, but communication delay increases due to long distance
Solution Approach 1:
The system segments the centralized cloud computing function into distributed edge computing nodes deployed at network access points. This segmentation allows tasks to be processed closer to users, reducing communication delay while maintaining computing capability through distributed processing across multiple edge servers.
Solution Approach 2:
The system transitions from a single-dimensional centralized cloud architecture to a multi-dimensional edge computing architecture that adds spatial distribution as a new dimension. Tasks can be routed to multiple edge servers based on network conditions, server loads, and task characteristics, enabling optimization along multiple dimensions simultaneously.
2Loss of time
If tasks are dispatched to edge servers dynamically, then response time can be reduced, but network congestion and server overload may occur
Solution Approach 1:
The system implements real-time feedback mechanisms where the online learning dispatcher continuously monitors network conditions and server loads, and adjusts task dispatching decisions accordingly. This feedback loop prevents network congestion and server overload by dynamically adapting dispatching policies based on current system state, thereby maintaining reliability while optimizing response time.
Solution Approach 2:
The system employs dynamic task dispatching and scheduling policies that adapt to changing network conditions and server loads in real-time. The online learning dispatcher and DRL-based scheduler continuously adjust their decisions based on current system state, enabling the system to optimize response time while preventing congestion and overload through dynamic adaptation.
3Productivity
If resources are allocated to minimize average response time, then efficiency improves, but scheduling fairness deteriorates causing task starvation
Solution Approach 1:
The system changes the decision parameters of the scheduler by incorporating fairness constraints and multiple objectives into the DRL-based scheduling policy. Instead of optimizing solely for average response time, the scheduler considers task waiting time, deadline requirements, and fairness metrics, adjusting resource allocation parameters to balance efficiency and fairness simultaneously.
Solution Approach 2:
The DRL-based scheduler is designed with multi-functionality to simultaneously optimize for multiple objectives including average response time, scheduling fairness, and task deadline satisfaction. This universal scheduler can adapt its behavior based on task characteristics and system state, providing both efficiency and fairness through a single integrated decision-making mechanism.
4Productivity
If online learning and deep reinforcement learning are combined, then dispatching and scheduling efficiency improves, but system complexity increases
Solution Approach 1:
The system segments the complex AI-based dispatching and scheduling function into two separate but coordinated modules: an online learning dispatcher for task-to-edge-server assignment and a DRL-based scheduler for resource allocation at each edge server. This segmentation reduces overall system complexity by dividing the decision-making process into manageable components while maintaining high efficiency through their coordinated operation.
Data Source
AI summary
The present disclosure relates to an online task dispatching and scheduling system. The system includes an end device; an access point (AP) configured to receive a task from the end device; one or more edge servers configured to receive the task from the AP, the one or more edge servers including a task waiting queue, a processing pool, a task completion queue, and a scheduler, wherein the AP further includes a dispatcher utilizing Online Learning (OL) for determining a real-time state of network conditions and server loads; and the AP selects a target edge server from the one or more edge servers to which the task is to be dispatched; and wherein the scheduler utilizes Deep Reinforcement Learning (DRL) in generating a task scheduling policy for the one or more edge servers.