READ: Robustness-Oriented Edge Application Deployment in Edge Computing Environment

In recent years, edge computing has emerged as a prospective distributed computing paradigmthat overcomes several limitations of cloud computing. In the edge computing environment, a service provider can deploy its application instances on edge servers at the edge of the network to serve its own userswith low latency. Given a limited budgetK for deploying applications on the edge servers in a particular geographical area, a number of approaches have been proposed very recently to determine the optimal deployment strategy that achieves various optimization objectives, e.g., tomaximize the servers' coverage, tominimize the average network latency, etc. However, the robustness of the services collectively delivered by the service provider's applications deployed on the edge servers has not been considered at all. This is a critical issue, especially in the highly distributed, dynamic and volatile edge computing environment. In this article, wemake the first attempt to tackle this challenge. Specifically, we formulate this Robustness-oriented Edge Application Deployment (READ) problemas a constrained optimization problemand prove its NP-hardness. Then, we provide an integer programming based approach named READ-O for solving this problemprecisely. We also provide an approximation algorithm, namely READ-A, for finding near-optimal solutions to largescaleREADproblems efficiently. We prove its approximation ratio is not worse than K/2, which is a constant regardless of the total number of edge servers. We evaluate our approaches experimentally on a widely-used real-world dataset against five representative approaches. The experiment results demonstrate that our approaches can solve the READproblemeffectively and efficiently.

Automated summary

This article introduces the concept and advantages of edge computing, and proposes the problem of robustness-oriented edge application deployment. Through mathematical modeling and algorithm design, precise and approximate solutions are provided to solve this problem. Experimental results demonstrate the effectiveness of the proposed methods on real-world datasets.