| With the exponential growth of network-based services,communication networks have been one of the mission critical infrastructures in our society.People increas-ingly rely on a variety of network services provided by the large-scale communication network,including business services,medical treatment,financial activities,etc.How-ever,network infrastructures are facing many potential threats,such as natural disasters and man-made destruction.These potential threats may result in multiple nodes or links failure simultaneously in a specific region,which is called regional failure.The region-al failure could cause many network services cease to function,and severely affects the user experience of communication networks.After a regional failure,how to quickly restore the disconnected network and min-imize the impact on the network services,is the critical issue for network operators to provide reliable networks.Conventional mechanisms to maintain network continuity can be briefly classified into two categories:protection and restoration.Protection is a proactive recovery approach,which relies on the pre-established backup resource to provide fast failure recovery.However,such proactive approach may fail to deal with regional failures since the backup resource might corrupt at the same time.Restora-tion mechanism computes new routes based on the states of current network.It is more flexible to exploit the diversity of network topology.However,the restoration process requires both unambiguous failure localization and sophisticated algorithms to discover alternate paths,both of which will typically slow down the recovery cycle.In this thesis,our connection recovery strategy combines the advantages of both protec-tion and restoration strategies,which ensures rapid connections recovery,and flexible network resources assignment.This thesis proposes a new connection recovery strategy based on overlay network and source routing.This strategy takes advantage of the idea of overlay network and single-hop source routing to exploit the network path diversity.Through a group of carefully selected immediate nodes(named as landmarks),the system can find a new transmission path for a failed connection.Without the knowledge of the location and scale of the region failure,the traffic can bypass the failure region with source routing.Extensive simulations show that our recovery system can improve the recovery ratio about 10%than the traditional strategies.Meanwhile,a very small fraction of land-marks(4 to 5 landmarks)is enough to achieve the recovery ratio of more than 90%.In order to make our recovery system more practical,we consider the network load when recovering failure connections.We design and implement a congestion-aware path recovery scheme based on the overlay network.With the hop-by-hop routing algorithm,the recovery paths can bypass both the failure region and congested nodes.In addition,the algorithm can also ensure that the traffic is transmitted along a loop-free path.In order to verify the feasibility of the system,we implement the recovery system by modifying and forwarding the packets in the IP layer based on the Linux Netfilter framework.The system consists of the client module and the overlay node module.The client module forwards the traffic to an appropriate overlay node once detecting the failure.Then the overlay node module acts as a NAT proxy to relay the traffic to the next hop through an less congestion path.We deploy the recovery system on the Emulab network testbed.The experiments on the real network show that the system can reset a failed connection within one second,and there are only introduce several KBits network probing overhead among overlay nodes.In addition,the system can support multiple transport protocols(including TCP and UDP).Finally,in order to identify the impact of region failure intuitively,this thesis first presents the concept of degree of network damage based on the seismic intensity scale.The degree of network damage is classified into five scales,which can measure the im-pact on the network infrastructures,network services from time and space dimensions.The degree of network damage provides a deep understanding of region failures and benefits the disaster-resilient network design,network vulnerability assessment as well as designing post-disaster recovery. |
