In the ongoing quest to develop a comprehensive Meta-OS for the continuum, the landscape is marked by existing concepts and frameworks aimed at unifying edge, cloud, and IoT resources.
Compatibility with diverse container management frameworks
In a computing world where virtualization of workloads is imperative, some companies still base on Docker management, while others have shifted to cloud-native environments, mostly relying on Kubernetes-only setups. Eclipse aeriOS, however, stands out by leveraging existing concepts to significantly extend the current state-of-the-art architecture. It advances orchestration and management capabilities beyond what is currently available, offering a more flexible, robust, scalable, and efficient solution for the modern Cloud-Edge-IoT landscape. It offers the capacity to execute loads in both environments, and opens up customization for future incorporations (e.g., containerd, Wasm-based).
Seem centralized, act distributed
Eclipse aeriOS overcomes the impression of vendor-agnosticity. Using a single interface, all resources are presented the same way for the user. They can be monitored and manipulated albeit residing in different networks, be owned by different companies or having disparate characteristics. Nonetheless, Eclipse aeriOS avoids centralization: the access to the information is ubiquitous and the orchestration decisions are taken in a decentralized manner. Thanks to balancing algorithms, the requests for workload commissioning (and other key processes) are handled in different spots avoiding single point of failure effects.
Standard-based communication and data management
The Meta-OS relies on using de-facto standard communication technologies, such as HTTP REST APIs, OpenAPI documentation, solid data formats such as NGSI-LD and tunnelling based on Fully-Qualified-Domain-Names for required networking (employing Wireguard VPNs). Also, due to inheriting from a research project, it aligns with impactful on-going initiatives, such as the TF3 Architecture from EUCEI.
What does Eclipse aeriOS provide?
It provides a series of components that are installed over a baseline infrastructure to be provided by the adopter (typically Kubernetes clusters, virtual machines, native systems or physical machines).
Federation of domains
Infrastructure Elements (IEs) are the fundamental computing units Eclipse aeriOS, providing a unified runtime environment. A group of IEs forms a domain, the smallest administrative entity, sharing core services. Domains connect to form a Eclipse aeriOS continuum, supporting a federated orchestration model facilitated by the Context Broker Orion-LD which implements NGSI-LD interface. This structure ensures peer-to-peer collaboration among domains, enabling autonomous, decentralized decision-making and fine-grained resource control.
As the Meta-OS builds on such concepts of Domains and Infrastructure Elements, it presents differences in the components to be installed based on the topology (that is decided by the IT administrator of the adopter entity).
Deployment of containerized workloads
Eclipse aeriOS introduces a two-layer orchestration model separating decision-making from execution. The High-Level Orchestrator (HLO) -a custom Python-based framework that employs Redpanda as messaging bus- acts as the decision engine, using global resource awareness to optimize workload placement across domains. It is formed of various modules, namely HLO Frontend, HLO Data Aggregator, HLO Allocator, HLO engine and HLO Local Allocation Manager. The Low-Level Orchestrator (LLO), which materializes in Go-based Kubernetes operators- handles local enforcement, translating HLO decisions into actionable commands on specific resources. This hierarchical design ensures scalability and adaptability without disrupting other domains.
Continuous observability of computing elements
To unify diverse and distributed resources, Eclipse aeriOS uses a Smart Data Model from project aerOS to abstract and pool them for dynamic workload execution. It gathers real-time data on resource capabilities and availability using Prometheus or customized scripts running on all Infrastructure Elements. This comprehensive view enables the Meta-OS to make efficient placement decisions and support proactive workload migration.
Unified management portal
Eclipse aeriOS is accessed via a web-based modern GUI that connects with the Meta-OS backend. It allows observing the computing resources and the deployed services, as well as to commission workloads (specifying the requirements) but also to compare the performance of such elements against relevant (edge, IoT) benchmarks.
Move decision and intelligence to the edge
Nodes are no longer passive elements that take orders and push monitoring/logging data. In Eclipse aeriOS, those are (depending on their capacities) able to trigger local orchestration notifications (e.g., to offload if saturated), scale horizontally, detect anomalies on data or on their behaviour, and to adapt the sampling frequency to certain circumstances.
Trustworthiness in the continuum
Eclipse aeriOS integrates cybersecurity services for robust authentication, authorization, and access control, ensuring secure access to domain resources through role-based policies and validated identity registries (using LDAP). Its trust management framework assesses the reliability of Infrastructure Elements via a Trust Agent and Trust Manager, using metrics like behavior, health, and reputation. Trust and reputation are key, with mechanisms to ensure message immutability and trust-based resource selection (via the usage of IOTA Tangle). Security is handled holistically, combining centralized IAM (Keycloak, KrakenD) with encrypted communications (TLS, VPN) and fine-grained data access control.
In addition, it includes the option of using Shapley weights to reinforce explainability of compatible ML models used over the Meta-OS.
Custom functions definition and deployments
Eclipse aeriOS embeds a customized version of OpenFaaS to allow the execution of one-shot, serverless applications on demand. Once installed (typically, in the most cloud-liked infrastructure of the continuum), the adopter may code and upload their own applications based on pre-defined templates, which could asynchronously trigger process such as ETA, data curation or statistics visualization in natively-equipped Grafana.
