AppSlice: A system for application-centric design of 5G and edge computing applications

TL;DR

AppSlice introduces a unified abstraction for edge computing and 5G networks that jointly manages compute and network resources, improving application performance and resource efficiency.

Contribution

The paper presents AppSlice, a novel declarative abstraction and runtime system that jointly manages compute and network resources for edge and 5G applications.

Findings

Significant performance improvements on a 5G testbed.

Effective joint management of compute and network resources.

Demonstrated benefits in real-time application scenarios.

Abstract

Applications that use edge computing and 5G to improve response times consume both compute and network resources. However, 5G networks manage only network resources without considering the application's compute requirements, and container orchestration frameworks manage only compute resources without considering the application's network requirements. We observe that there is a complex coupling between an application's compute and network usage, which can be leveraged to improve application performance and resource utilization. We propose a new, declarative abstraction called AppSlice that jointly considers the application's compute and network requirements. This abstraction leverages container management systems to manage edge computing resources, and 5G network stacks to manage network resources, while the joint consideration of coupling between compute and network usage is explicitly managed by a new runtime system, which delivers the declarative semantics of the app slice. The runtime system also jointly manages the edge compute and network resource usage automatically across different edge computing environments and 5G networks by using two adaptive algorithms. We implement a complex, real-world, real-time monitoring application using the proposed app slice abstraction, and demonstrate on a private 5G/LTE testbed that the proposed runtime system significantly improves the application performance and resource usage when compared with the case where the coupling between the compute and network resource usage is ignored.