On the trade-off between cost and availability of virtual networks
2014
To minimize cost, Virtual Network Operators (VNOs) need to consider the required network availability already at the network design stage. One generic approach to reach the availability target is to select only high-quality physical network elements that offer high availability and consequently demand high expenses per element. The other generic approach to achieve high availability is to add protection capacity on the level of the virtual network based on lower cost components. In this paper, we analyze both alternatives with a simulation tool to answer the fundamental question how quality can be traded against capacity. For this purpose, we consider different network topologies and the influence of different parameters and provide a framework to find an optimal strategy between Mean Time Between Failures (MTBF) targets for the physical infrastructure and the usage of additional backup paths on the virtual network level. I. INTRODUCTION Besides connectivity and capacity, any carrier-grade vir- tual network has also to comply to availability targets at coping with fiber cuts and other failures. Constantly trying to minimize cost at renting (virtualized) links and nodes from Physical Infrastructure Providers (PIP) (1), the Virtual Network Operator (VNO) therefore faces a basic choice: it may build upon highly reliable network elements (nodes and links) or apply protection and restoration mechanisms on the basis of a larger number of network elements with lower availability figures (and thus lower cost). The first option - the 'high cost physical network' approach - uses direct, shortest paths with high availability basing on high cost links, i.e. the operator will invest in the infrastruc- ture. The second option - the 'low cost physical network' approach - realizes the necessary network availability in the virtual network domain by combining several parallel paths with lower availability and lower individual cost. Here, the individual physical network elements can be kept cheap, while a larger number of them are required to realize the parallel paths. Thus, a trade-off can be expected between link quality (small number of expensive paths) and capacity (combination of multiple cheap paths). In this paper, we examine this trade-off between the two choices of a 'high cost physical network' approach and a 'low cost physical network' approach focusing on the optical links. Our contribution is a framework to identify the cost optimal values of the Mean Time Between Failures (MTBF) parame- ters of the physical links considering the influence of different parameters and network topologies. Further we analyze the fundamental inter-dependencies and give recommendations. To the best of our knowledge this is the first paper to consider how changes in the underlying physical infrastructure (higher MTBF values or usage of several backup paths) will interact to achieve the desired link availability of a Virtual Network Request (VNR) - the embedding with lowest cost. In the following sections, we first consider the (fiber) availability and how to determine it from measured MTBF values. Further, we model the relationship between MTBF and cost derived from existing values. Next, we explain our virtual network embedding algorithm that is mapping a VNR with desired availability onto the physical network while mapping the virtual nodes to the matching physical nodes and the virtual link to a single or several parallel physical (backup) paths. This algorithm is then used to investigate the physical infrastructure deployment strategy considering the MTBF to achieve cost reduction. Several cost-model parameters, different network topologies and sizes, and requested link availability values will be considered. We show that the network topology (nodal degree) and size is strongly influencing the Virtual Network Embedding (VNE) deployment strategy to achieve minimum cost. In the last section we give some recommendations for network operators derived from our results.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
24
References
8
Citations
NaN
KQI