Abstract:With the development of IoT and 5G technologies, multi-access edge computing has been widely concerned due to the advantage of providing extra computation resource in edge servers for end-devices to complete their computation tasks with low latency and great reliability. Although considerable efforts on tasks offloading have been developed to reduce the energy consumption and computation latency, most of these efforts focused on independent tasks offloading in which the tasks ofÂ end-devices that are offloaded to edge servers are mutually independent, and offloading dependent tasks in which the execution of these tasks has strict dependencies is still a challenge and need to be further investigated. To this end, this paper considers the multi-access edge computing with channel interference and proposes a deep reinforcement learning based dependent task offloading scheme, namely DRL-DTO, to effectively offload dependent tasks of end-devices to edge servers for tasks computation with low energy consumption and low latency. Particularly, the directed acyclic graph (DAG) is introduced in our scheme to conduct the dependency model of dependent tasks, and a dependent task offloading model is developed with considering multiple constraints of task dependency, communication channel quality, computation latency and energy consumption. Based on the DAG model and dependent task offloading model, a priority queue for tasks of various end-devices is conducted and the dependent task offloading is transferred as a markov decision process (MDP). Then, a deep reinforcement learning based task offloading approach is proposed to solve the MDP of dependent task offloading and determine the effective offloading decision with objective of minimizing both the energy consumption of end-devices and computation latency of offloaded dependent tasks. Through performance evaluations, our results demonstrate that, in various scenarios including various property of edge-servers, various number of dependent tasks and various DAG dependency of dependent tasks, various preference factor, our DRL-DTO scheme can always achieve lower computation latency and energy consumption than existing schemes.

Offloading Dependent Tasks in Edge Computing with Unknown System-Side Information

Offloading Decision Methods for Multiple Users with Structured Tasks in Edge Computing for Smart Cities

Decentralized Task Offloading in Edge Computing: A Multi-User Multi-Armed Bandit Approach

Dynamic Task Offloading in Edge Computing based on Dependency-aware Reinforcement Learning

Dependent Application Offloading in Edge Computing

Exploiting Function-level Dependencies for Task Offloading in Edge Computing

Efficient Task Offloading with Dependency Guarantees in Ultra-Dense Edge Networks

A Novel Deep Reinforcement Learning Based Dependent Tasks Offloading in Multi-Access Edge Computing

Dependency-Aware and Latency-Optimal Computation Offloading for Multi-User Edge Computing Networks

Online Computation Offloading for Deadline-Aware Tasks in Edge Computing

Dependent Task-Offloading Strategy Based on Deep Reinforcement Learning in Mobile Edge Computing

Real-Time Offloading for Dependent and Parallel Tasks in Cloud-Edge Environments Using Deep Reinforcement Learning

Dependent Task Offloading for Edge Computing based on Deep Reinforcement Learning

TODG: Distributed Task Offloading with Delay Guarantees for Edge Computing

Offloading Tasks to Unknown Edge Servers: A Contextual Multi-Armed Bandit Approach.

A Near-Optimal Approach for Online Task Offloading and Resource Allocation in Edge-Cloud Orchestrated Computing

Optimization Search Strategy for Task Offloading from Collaborative Edge Computing

Collaborative Offloading Strategy for Dependent Tasks in Mobile Edge Computing

Dependency-aware task offloading based on deep reinforcement learning in mobile edge computing networks

Dependency-aware task cooperative offloading on edge servers interconnected by metro optical networks

Deep reinforcement learning-based online task offloading in mobile edge computing networks