Distributed service migration method for mobile edge computing

Abstract

The invention discloses a distributed service migration method for mobile edge computing, and belongs to the field of Internet of Things. According to the method, the mobile management problem in mobile edge computing MEC is researched. When a device is in a moving state, computing tasks need to be dynamically migrated among a plurality of edge servers to maintain continuity of services. Due to uncertainty of movement, cost and delay can be increased due to frequent migration, and service interruption can be caused due to non-migration. Therefore, in multi-agent deep reinforcement learning MADRL, an adaptive weight depth deterministic policy gradient AWDDPG algorithm is introduced to optimize the cost and delay of multi-user task migration, and a centralized training and distributed execution method is used to solve the high dimension problem during task migration. A large number of experiments show that compared with related algorithms, the algorithm provided by the invention greatly reduces service delay and migration cost.

Description

technical field [0001] The invention belongs to the field of the Internet of Things, and in particular relates to a distributed service migration method oriented to mobile edge computing. Background technique [0002] In recent years, with the continuous development of big data, artificial intelligence, Internet of Things (IoT), MEC and other technologies, mobile devices have become more and more widely used in people's lives, such as VR, AR, smart home and so on. These devices usually have computation-intensive and latency-sensitive tasks, but the limited resources of mobile devices are difficult to meet the above application requirements. The traditional solution is to offload computing tasks to cloud centers with sufficient resources for processing, but long-distance transmission will increase computing delays. MEC is to deploy servers at the edge of the network. The edge servers are geographically closer to users, which can effectively reduce service delays. Resource-c...

AI Technical Summary

Problems solved by technology

Some traditional methods transfer tasks by predicting the user's location, but it is difficult to predict the user's mobility in practical application scenarios

There is also the application of deep Q-learning (DQN) to task migration. Although DQN can handle complex state spaces, the centralized processing method cannot meet the task migration requirements of multi-user edge computing. As the number of users increases, the state space of the system and the dimensionality of the action space will grow exponentially

And in a multi-user scenario, the states of all users are combined into a global state, resulting in an unstable multi-user environment, ignoring the impact between users

Distributed deep reinforcement learning can effectively solve the above problems, but in a distributed environment, each agent makes decisions independently and cannot ignore the interaction with other agents in the environment.

Combining the local states of all mobile users into a global state for training can solve the problem of interaction between the agent and the environment. State will not only lead to instability in multi-user environment, but also frequent interaction will increase the communication cost

Therefore, it is very challenging to design an effective transfer strategy based on distributed deep reinforcement learning to balance transfer cost and latency.

Images