Conceptes and Methods for Edge Computing
Final Report Abstract
A strong increase in sophisticated applications is expected for the Internet, for example in the areas of augmented reality (AR) and autonomous driving. This poses major challenges for Internet core networks and the prevailing paradigm of cloud computing. To enable latency-critical applications that process large amounts of data, edge computing attempts to place computing/storage resources called cloudlets closer to the data sources. In the SmartEdge project, concepts and methods for edge computing were researched. The emphasis was put on urban areas, where applications of AR, autonomous driving etc. entail special requirements. First, the physical infrastructure for cloudlets in urban areas was investigated. Concepts, analysis and planning tools were developed for upgrading existing infrastructures in urban areas - cellular base stations, WLAN routers and streetlights - with cloudlets. An empirical study analyzed various metrics for coverage. To our knowledge, this is the rst large-scale study of potential cloudlet coverage in urban areas based on real-world datasets. In addition, a placement algorithm was developed to determine cloudlet placement based on local requirements to best meet the computational needs of mobile applications while minimizing cost. For service provisioning, a novel edge service execution environment was developed based on microservices—lightweight computation units that provide a speci c functionality. A new approach to o oading computations reduced the end-to-end latency for service execution. At the same time, the developed approach signi cantly reduces the energy consumption of the clients. This is particularly relevant because battery-powered mobile devices are often involved in the application scenario. Various control and adaptation mechanisms were developed for the execution environment. Speci cally, both theoretical foundations and fast algorithms were developed to enable e cient management of edge platforms. For service discovery, a scalable approach was developed that incorporates known protocols and applications. To enable elasticity in execution, a concept was developed to make microservices adaptable. In this way, lower resource requirements can be achieved if losses in the quality of computation are acceptable. Finally, an extensive literature review was conducted to investigate the characteristics and bene ts of edge computing for a variety of application domains.
