About EDGELESS

The goal of EDGELESS is to leverage the serverless concept in all the layers in the edge-cloud continuum to fully benefit from diverse and decentralised computational resources available on-demand close to where data are produced or consumed. In particular, we aim at realising an efficient and transparent horizontal pooling of the resources on edge nodes with constrained capabilities or specialised hardware, smoothly integrated with cloud resources, which is a giant leap forward compared to state-of-the-art vertical offloading solutions where the edge is a mere supplement of the cloud.

Progress beyond state of the art. In EDGELESS, we will explore the dynamic configuration and adaptation of distributed ML/AI algorithms based on (i) the (limited) resources of heterogeneous sets of edge devices; (ii) constraints imposed by applications; (iii) variable networking environments and performance.

First, we plan to leverage AI algorithms for forecasting not only
network traffic, but also user and workloads mobility, anomalies or business constrains to cite few examples; such aspects are not considered by existing work. All this information can be used as input for orchestration, resource scheduling and load balancing.

For this purpose, we will use a mixture of real datasets collected from operational networks of project partners and also synthetic datasets from lab environment and controlled phenomena. We will also explore the use of deep learning for resource orchestration and task offloading within the EDGELESS framework that considers the dynamic fluctuations and patterns in network topologies and user requests to tailor the location and instances of application tasks offloaded in the edge. With specific reference to an anomaly detection function, we will generate controlled anomalous events and capture these in the datasets, to avoid running the system for a very long time to have a sufficient number of events.


Second, we will extend existing solutions to deal with heterogeneity of devices and distributed environment with small function granularity. The challenge will be to understand the necessary information and specify the requirements in terms of metrics that can be extracted from diverse
environments, e.g., device and resource types.

In EDGELESS we aim at realising horizontal offloading across the continuum, from devices to the cloud, to maintain the edge promise (low latency, high bandwidth, high reliability) everywhere.
This will result in the ability to deploy serverless workflows in a far–/near–edge system as swiftly as in the cloud, while keeping the state close to compute elements to support data–intensive applications and seamlessly managing local physical resources to optimise the environmental footprint.

EDGELESS framework components (ε–orchestrator, ε–controller, ε–balancer) will incorporate cognitive functions to dynamically adjust the execution of the functions to support user mobility and network dynamicities. Hardware diversity will be exploited to maximise performance by dynamically allocating lambda executors and ancillary services on edge nodes, using lightweight virtualisation abstractions and trusted execution environments when available, under fast changing load conditions.

Development and validation of EDGELESS’s technical features will follow an agile methodology (in a broad sense): they will be organised around iterative cycles of design, modelling, development, integration, and experimentation. Nonetheless, for the purposes of project management and to support external reviews, we define a release plan, aligned to the deliverables and milestones defined in the work plan .

Software development is a central activity, since it bridges the abstract design of architecture, interfaces, and algorithms to the more practical validation activities.

Integration is intended as a continuous circular process, starting with an initial preparation phase to set up all the tools required for an effective and efficient collaboration between the multiple contributing partners.

Three pilots will provide a realistic environment for respective high–impact use cases where to
validate the overall EDGELESS system with real applications in a close–to–production system:

  • Autonomous Smart City Surveillance (ASCS)
  • Internet of Robotic Things (IoRT)
  • HealthCare Assistant (HCA).

Three testbeds will be used to validate specific sets of features in controlled conditions, in environments that are representative for the given features under test, respectively:

  • 5G/MEC: features deployment on telco near–edge; inter–cluster / cloud resource scheduling
  • wireless edge: features efficient clustering in large–scale deployments of constrained edge nodes
  • distributed consensus: features as an orchestration tool and as a function library for EDGELESS applications
Scroll to Top