No abstract

A set of new measures for network structural dependency analysis is introduced. These measures are based on geodesic distance, which is the number of links in a shortest path. They capture the structural dependency effect at the path level, the node level, and the overall network level, and hence can be used to index such dependencies. Unlike the related literature measures, a novel aspect of the proposed measures is that the impact of network fragmentation caused by a node failure is taken into explicit consideration in deciding the structural dependency effect. As a result, when applied to critical node identification in a network, the proposed measures give results that are more in line with intuition.

Network Function Virtualization (NFV) is an emerging technology that reduces cost and brings flexibility in the provisioning of services. NFV-based networks are expected to be able to provide carrier-grade services, which require high availability. One of the challenges for achieving high availability is that the commodity servers used in NFV are more error prone than the purpose-built hardware. The "de-facto" technique for fault tolerance is redundancy. However, unless planned carefully, structural dependencies among network nodes could result in correlated node unavailabilities that undermine the effect of redundancy. In this paper, we address the challenge of developing a redundancy resource allocation scheme that takes into account correlated unavailabilities caused by network structural dependencies. The proposed scheme consist of two parts. In the first part, we propose an algorithm to identify nodes that can be highly affected by a node failure because of their network structural dependency with this node. The algorithm analyzes such dependencies using a recently proposed centrality measure called dependency index. In the second part, a redundancy resource allocation scheme that places backup network functions on nodes considering their dependency nature and assigns the instances to flows optimally is proposed. The results show that not considering the network structural dependency in backup placement may significantly affect the service availability to flows. The results also give insights into the trade-off between cost and performance.

No abstract

Network Function Virtualization (NFV) implements network middlebox functions in software, enabling them to be more flexible and dynamic. NFV resource allocation methods can exploit the capabilities of virtualization to dynamically instantiate network functions (NFs) to adapt to network conditions and demand. Deploying NFs requires decisions for both NF placement and routing of flows through these NFs in accordance with the required sequence of NFs that process each flow. The challenge in developing NFV resource allocation schemes is the need to manage the dependency between flow-level (routing) and network-level (placement) decisions.We model the NFV resource allocation problem as a multi-objective mixed integer linear programming problem, solving both flow-level and network-level decisions simultaneously. The optimal solution is capable of providing placement and routing decisions at a small scale. Based on the learnings from the optimal solution, we develop ClusPR, a heuristic solution that can scale to larger, more practical network environments supporting a larger number of flows. By elegantly capturing the dependency between flow routing and NF placement, ClusPR strikes a balance between minimizing path stretch and maximizing network utilization. Our experiments show ClusPR is capable of achieving near-optimal solution for a large sized network, in an acceptable time. Compared to state-of-theart approaches, ClusPR is able to decrease the average normalized delay by a factor of 1.2 − 1.6× and the worstcase delay by 9 − 10×, with the same or slightly better network utilization.