.png){kind=icon}
.png){kind=icon}
Return
Delay-Optimal Computation Offloading in Large-Scale Multi-Access Edge Computing Using Mean Field Game
.png){kind=icon}
PRE
AI
DOI：10.1109/TWC.2023.3291198.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In large-scale multi-access edge computing (MEC) networks, each device should make the computation offloading decision distributively. In this paper, we target on a delay-optimal computation offloading problem in large-scale MEC systems, where each task has two properties: data size and computation amount. Because the detailed state information of massive devices are huge in large-scale systems, we propose a distributed computation offloading algorithm using the mean field game (MFG). To design the distributed computation offloading algorithm, we first formulate the delay-optimal computation offloading problem as a Markov decision process (MDP) and derive the Hamilton-Jaccobi-Bellman (HJB) equation with the unknown task allocation proportion, where the combined influence from other devices and MEC servers should be estimated. Based on MFG, we obtain the Fokker-Planck-Kolmogorov (FPK) equation to describe the evolution of the system's collective behavior, with the influence from other devices and MEC servers formulated as the mean field. To solve the large-scale problem with the unknown allocation proportion, we propose a optimal computation offloading algorithm based on the generative adversarial networks (GAN) structure. For the generator, we generate the unknown task allocation proportion due to its non-calculability and insufficient dataset. For the discriminator, we train the value function, and propose a water-filling algorithm to prioritize the task offloading. Finally, the simulation results evaluate the performance of the proposed algorithm and show the performance gain compared to conventional algorithms.
Keywords:
Multi-access edge computing (MEC)
mean field game (MFG)
generative adversarial networks (GAN)
Journal
IF:
10.7
Papers:
1.3W
Citations:
5.3W