Performance and dependability evaluation in a convergente network service using BGP and BFD protocols

Abstract

Critical network services require maximum availability. The design of a convergent network must be oriented so that downtime is as short as possible. It is important to highlight that the term convergent refers to recovery in case of failure and not to the integration of data and voice in the same channel. For this reason, a network device must be able to quickly detect any communication failure between adjacent devices, so that the upper layer protocol can fix that failure and prevent any interruptions on the provided services. The failure detection time is dependent on issues intrinsic to the protocol used. The association between BFD (Bidirectional Forwarding Detection) and BGP (Border Gateway Protocol) protocols allows BFD to quickly detect failures in the connections between BGP peers, implementing fast convergence of BGP routes. This dissertation proposes two approaches for supporting the estimation of convergent architectures’ availability by considering different configurations and the related impact of each one on the performance. We have proposed a Continuous-Time Markov Chain (CTMC) model to represent network architectures using both protocols BGP and BFD. This CTMC represents architectures running critical systems and must be used to predict availability by considering a given configuration. Thus, our second approach is aimed at supporting an inferential performance evaluation by considering a paired before-and-after comparison. This approach is applied in order to obtain the best convergence condition for a corporate network and, as a consequence, to offer greater availability as possible. More specifically, this work proposes an approach to estimate the performance of the concentrating equipment, that supports the availability solution, by applying a model based on an M/M/1/K queue, which was changed in its original definition for supporting the analysis of BFD control packets. Sensitivity analyses were carried out by considering the mean arrival time (MAT) and discard rate of BFD control packets and the concentration equipment’s CPU utilization. We also present a sensitivity analysis by considering the network’s availability as a result of the fail-over time. In addition, we presented a closed-form equation to calculate the availability of critical warm-standby components for large network architectures. By combining all approaches, it is possible to evaluate the trade-off between availability and performance cost for implementing a solution and find out a configuration that offers the best solution by considering the project’s requirements. Maybe, in some circumstances, our strategies need to be refined in order to support different characteristics of the network architecture to be evaluated. Our strategy may be refined in order to support different characteristics of the network architecture. We proposed two methodologies for supporting the applications of our strategies. Three case studies were performed to evaluate the effectiveness of our approaches. Our approaches have proven to be feasible, and they highlight the most appropriate scenarios, supporting network architects.