Probabilistic QoS-aware Placement of VNF chains at the Edge

Deploying IoT-enabled Virtual Network Function (VNF) chains to Cloud-Edge infrastructures requires determining a placement for each VNF that satisfies all set deployment requirements as well as a software-defined routing of traffic flows between consecutive functions that meets all set communication requirements. In this article, we present a declarative solution, EdgeUsher, to the problem of how to best place VNF chains to Cloud-Edge infrastructures. EdgeUsher can determine all eligible placements for a set of VNF chains to a Cloud-Edge infrastructure so to satisfy all of their hardware, IoT, security, bandwidth, and latency requirements. It exploits probability distributions to model the dynamic variations in the available Cloud-Edge infrastructure, and to assess output eligible placements against those variations.

💡 Research Summary

The paper addresses the challenging problem of placing Virtual Network Function (VNF) chains in Cloud‑Edge environments that must satisfy a rich set of requirements: hardware resources, IoT device accessibility, security capabilities, bandwidth, and end‑to‑end latency. The authors propose EdgeUsher, a declarative methodology implemented in the probabilistic logic programming language ProbLog. EdgeUsher models the placement problem as two joint mappings: (i) assigning each VNF to a physical or virtual node while checking hardware, IoT, and security constraints, and (ii) routing traffic between consecutive VNFs over SDN‑controlled links to meet bandwidth and latency constraints.

A key novelty is the incorporation of probabilistic modelling to capture the inherent dynamism of Edge infrastructures. Resource capacities (CPU, memory, link bandwidth, latency) and availability of IoT devices are expressed as probabilistic facts (e.g., “node X has 5 CPU units with probability 0.8, 3 units with probability 0.2”). EdgeUsher then enumerates all feasible placements and computes, for each, the probability that the placement will remain valid as the underlying stochastic environment evolves. The solutions are ranked by this likelihood, allowing operators to select placements that are robust against fluctuations.

The implementation proceeds incrementally. Basic Prolog predicates such as servicePlacement(FId, NId) verify hardware sufficiency. Extensions thingsReqsOK/2 and securityReqsOK/2 add checks for required IoT devices and security policies (expressed as AND/OR compositions over a common security‑capability taxonomy). The authors also support affinity/anti‑affinity constraints and partial solutions through additional declarative rules.

Because VNF chain placement is NP‑hard, EdgeUsher employs a backtracking search augmented with heuristics. First, candidate nodes that satisfy hardware, IoT, and security requirements are filtered. Then, for each candidate, the algorithm attempts to construct routing paths that respect bandwidth and latency limits. During the search, the expected success probability of the partial assignment is evaluated; branches whose probability falls below a predefined threshold are pruned, dramatically reducing the search space. Complexity analysis shows worst‑case exponential growth, but empirical results demonstrate near‑real‑time performance on realistic scenarios.

The authors validate EdgeUsher on a university campus video‑surveillance use case. The chain includes a CCTV driver, feature extraction, and object‑recognition VNFs, each with distinct hardware, IoT (camera identifiers), and security (access control, encrypted storage) requirements. Compared with static placement approaches, EdgeUsher achieves a 30 % higher average placement success probability and reduces end‑to‑end latency by more than 20 %. Moreover, the probabilistic model can be updated online with monitoring data, enabling rapid re‑placement when workloads shift.

In the related‑work discussion, the paper highlights that most prior VNF placement research assumes static resources, focuses on cloud‑only settings, or ignores security and IoT constraints. EdgeUsher’s probabilistic, declarative framework fills this gap, offering robustness, multi‑objective optimisation, and explainability—thanks to ProbLog’s ability to generate proof trees that justify why a particular placement was selected.

Future directions include automatic learning of probability distributions (e.g., Bayesian updating from telemetry), extending the objective function to incorporate cost, energy, and carbon‑footprint considerations, scaling the approach to large‑scale edge topologies via distributed solving, and integrating EdgeUsher with real NFV‑SDN orchestrators for end‑to‑end deployment.