Robust embedding of VNF/service chains with delay bounds

The efficient and carrier-grade operation of virtualised network infrastructures (Infrastructure as a Service, IaaS) within Cloud Systems requires powerful methods for dynamic resource provisioning, virtual network functions (VNF) placement and interconnection. In the scientific literature, already several contributions related to the virtual network embedding (VNE) problem can be found, see [1] and the references therein as well as our previous contributions [2], [3]. Typically, the physical substrate infrastructure (network nodes with switching, processing and storage resources, and links with defined bandwidth) as well as the traffic demands of the virtual networks are given and the target is to minimise the embedding cost wrt. performance and QoS constraints (e.g. bandwidth guarantees, latency bounds). In this contribution, we propose a novel optimisation model based on the concept of Γ-robustness [4], [5] to deal with uncertainties in the traffic demand and as a consequence in the resource requirements of the VNFs while fulfilling individual average roundtrip delay bounds for each chain of VNFs. The Γ-robust optimisation model is formulated as a mixed-integer linear program (MILP). Moreover, in order to enhance the scalability of the model, a modified MIP-based Variable Neighbourhood Search (VNS) heuristic is proposed. The performance of the novel optimisation model and the heuristic is evaluated for different performance scenarios using a network topology example taken from SNDlib [6].