Zero Touch Predictive Orchestration: Automating Time-Series Models for the Cloud-Edge Continuum

The Cloud-Edge Continuum (CEC) assists in the implementation of latency-sensitive applications by allocating processing power near the data origin, especially at the far edge. However, these environments can change rapidly, so Zero Touch Management must be implemented, relying on time-series predictions. The ‘cold start’ issue represents a major challenge, as newly added nodes have no data to create predictive models for the node, and localized models do not generalize well to differing hardware and microservices operational characteristics. The aim of this paper is to describe an innovative fully automated time-series prediction framework, which is based on a new data-mixing approach. A simple, technology-neutral Resource Exposer (RE) at the infrastructure layer

is used to expose nodes dynamically and collect telemetry data. This data can be customized according to different parameters such as computation, network, and energy consumption. To remedy the infrequent local data samples from new nodes, the framework merges these samples into TimeTrack, a high-resolution, publicly available dataset. This merging utilizes the primary temporal structures of TimeTrack in conjunction with the local node data's calibration. The fused data is then fed into a Neural Architecture Search (NAS) engine, which is capable of generating highly sophisticated predictive models. This method of data merging is the main contributor to overcoming the cold start problem, resulting in significantly improved forecasting for the Cloud-Edge

Continuum.

Message: Hi FE, I need help with “Zero Touch Predictive Orchestration: Automating Time-Series Models for the Cloud-Edge Continuum” in “Computer Science & Engineering”