There was a Linux Foundation Edge Akraino face-to-face TSC and 2020 planning meeting last week held at Facebook in Menlo Park, CA. It was a fascinating meeting where I got to see the state-of-the-art of edge computing; here are my key takeaways:
1) The dominance of Kubernetes (K8s): I think it might be safe to say that K8s will be the default NFVI on the edge. Most blueprints (the term used by Akraino to signify projects) were based on K8s. The few that were based on OpenStack used a containerized version of OpenStack deployed on K8s. So essentially it is safe to say that K8s will dominate edge environments.
2) Focus on a small footprint: Edge environments could be quite constrained. There were several blueprints focusing on very small environments such as on-prem. Specifically, see slide presentations on the ELIOT (Edge Lightweight and IOT), uMEC (Micro MEC), KubeEdge, and others.
3) Centralized infra controller: With 100s to 100s of thousand edge locations, it is impractical to install K8s on each edge manually. Almost every blueprint had some notion of a centralized infrastructure controller that would automagically install K8s on a given edge cloud. See IEC and ICN blueprints as example.
4) Diversity of applications: Another interesting aspect was the diversity of applications. Apps ranged from LTE, 5G, SD-WAN, AR/VR, gaming, V2X, IoT, and more. The edge is indeed going to a lot more than just network functions.
5) Framework wars: From general frameworks, e.g. TARS to specialized frameworks, e.g. Kubeflow for AI/ML apps, I expect numerous edge application frameworks to jockey for position. I expect full blown edge application framework wars to emerge in earnest in late 2020. These wars will be critical to establish control points, similar to IOS and Android for phones.
6) Interest in dataplane acceleration: RAN and UPF network functions are expected to put heavy demands on datapath computation. For this reason, hardware acceleration is expected to play an important role on the edge. Dataplane acceleration will come in numerous flavors ranging from software techniques, SmartNICs, FPGAs, GPUs, inference engines, and more.
Overall, I truly enjoyed the event and believe Akraino will play a key role in moving the industry forward. Aarna is involved with the Akraino ICN Blueprint helping with ONAP4K8s from an orchestration, lifecycle management, and automation point of view. Contact us if you want to learn more.
