| dc.description.abstract | Ubiquitous wireless communications have led to an increase in the importance of mobile networks in disaster recovery and mission-critical services. Network design and operation of mobile network infrastructure requires survivability as a fundamental requirement. Therefore, a quantifiable approach to survivability analysis of such infrastructures is crucial. The objective of this thesis is to propose analytical models for quantifying the survivability of infrastructure-based wireless networks subject to massive failures caused by natural disasters. This research work mainly focuses on the availability of network connectivity service, which is one of the major concerns of network operators for various disaster-based failures.
First, the survivability of a two-tier infrastructure-based wireless network in the presence of disastrous failures is modeled and analyzed. This thesis defines survivability as the network connectivity from the beginning of the failure until the system is fully recovered. The transient behavior of the studied network after a large breakdown is characterized using a Markov model. In addition, two modeling methods are proposed in this thesis, namely the exact model and the approximate product-form model. The results show that the approximate product form method provides near accurate results with precise lower computational complexity.
Second, network performance modeling needs to consider the location of wireless base stations and users. To this end, stochastic geometry is used to estimate spatial average network performance and integrate it into the survivability modeling of a two-tier infrastructure-based wireless network. In order to avoid the explosion of state space while addressing large networks, an approximate product-form analysis method is proposed, which decouples the two tiers so that the survivability analysis can be studied independently. In addition, the assumptions used in the proposed models are validated on the actual data, including the Poisson point process and product-form decomposition. Numerical experiments are also performed to examine the effects of different parameters on network survivability.
Another aspect that is covered by this work is the correlated failures caused by disastrous events. In particular, natural disasters can spread across geographical areas in a short period of time, and the resulting network failures can occur in multiple locations. Therefore, a multi-phase recovery model is developed in this thesis to quantitatively assess network survivability under fault propagation caused by disasters. Based on the model, the network survivability analysis is exemplified for three repair strategies. The results show that by carefully selecting the repair strategy, network performance loss caused by disasters can be reduced.
Moreover, in most literature the state holding time is assumed to be exponentially distributed, which may not be true in reality. To solve this problem, this work develops a non-Markovian survivability assessment model and conducts the survivability analysis based on the proposed model. The phase-type (PH) distribution technique is applied to approximate the nonexponential distributions, resulting in a Markov structure to simplify the analysis and obtain tractable analytical results. A numerical investigation is conducted to validate the PH approximation used in the proposed model. A case study illustrates the effect of different model parameters on network survivability.
The results of the above modeling analysis in this research work shed new insights not only on survivability analysis but also on survivability provisioning, e.g. how the model parameters affect the network’s survivabilityof such a network against failure events. Operators can use these quantification results to assess the trade-offs between survivability network design and survivability performance. Based on such trade-offs, operators can further implement optimization algorithms, such as providing network connectivity with minimal cost. Applying survivability analysis as part of an optimization framework to support network investment could be left for future work. | |
