Today’s data networks and cloud-computing infrastructure are highly optimized toward common applications with downlink-dominant network traffic characteristics, such as web-browsing and video- and music streaming, that prioritize bandwidth over responsiveness. However, they are unsuitable for novel emerging closed-loop applications, such as networked control systems or immersive AR, which have completely different traffic characteristics and require extremely low round-trip latencies. This has led to the emergence of a new distributed computing paradigm which aims to deploy networked computing resources as close to the application as possible. Known as Edge Computing, this paradigm is quickly becoming reality together with new networking technologies (5G, beyond-5G) which further benefit these highly latency-constrained applications.

However, while it is widely understood that Edge Computing brings much-needed improvements in performance, little is known yet about the detailed characteristics of these systems and their implications for applications. Essentially, there is a gap between theoretical work and the achievable performance of edge computing systems in practice, a gap that must be bridged before these systems can become a widespread reality.

Our goal with the ExPECA testbed is to bridge this gap. We provide a cluster of hardware-reconfigurable general-purpose computing nodes interconnected using managed switches and Software-Defined Radios. This allows us to quickly, on-the-fly, and in an automated fashion change the characteristics of the cluster and the network, in order to study different Edge- and Cloud-computing deployments and the applications that run on them.

Ainur is a framework for wireless testbed automation with a specific focus on end-to-end experimental research in the context of edge-computing using cloud- and edge-native technologies. Ainur is designed to deploy experimental runs from a workload perspective by configuring the physical testbed, initializing all involved software components, deploying and executing the experimental workload, collecting logs and data, and finally gracefully degrading the system. The framework allows for dynamic, software-definition of physical and logical links, network topology, cloud and edge computing resources, as well as experimental workload deployment and orchestration.