|
Online Optimization for Edge Computing The proliferation of latency-sensitive mobile applications is propelling the telecommunication industry towards embracing the paradigm of mobile edge computing (MEC), where low-latency computation services are provisioned at the network edge. In an MEC network, edge devices within a local area form a shared resource pool to create virtualized environments for task execution, which presents several challenges in designing an offloading-scheduling mechanism: 1) Requests are generated online, compelling the system to make irrevocable decisions under incomplete information. 2) Resources at the edge is highly constrained, both the sharable (memory) and non-sharable (CPU and bandwidth) resources require judicious management to maximize efficiency. 3) Emerging services such as AR/VR and industrial control impose strict deadlines for task completion. This necessitates consideration of user interference in the wireless transmission environment and the temporal and spatial sharing of computing resources.
We proposed a competitive algorithm for task scheduling in MEC network, where the channel assignment, function dispatching and the temporal management of the computing resources are optimized online. We analyzed the worst-case performance and prove its competitive ratio in the case of monolithic task scheduling, and proposed an efficient algorithm to schedule chains of dependent virtual functions.
|
