Distributed Failure Recovery of the LDP Signalling Protocol

Abstract

After a control plane failure, a signalling protocol needs to recover state information about the established connections in the data plane, so that new connection setups do not disrupt existing connections, and existing connections are not degraded (e.g., loss of connection information). First, we provide a comparison of existing techniques to achieve label distribution protocol (LDP) recovery. Then, we propose a backup mechanism to store state information in an upstream neighbour node. The backup LDP state information is synchronized with the original LDP state information in a downstream node when the LDP sets up or tears down connections. After that, we present a two-step LDP state information recovery, which uses a fast LDP state information recovery to recover what labels are idle before a control plane failure, and a detailed LDP state information recovery to fully recover all LDP state information. The fast LDP state information recovery is realized as part of the LDP initialization, allowing a restarting LDP session to process new connection setup requests as soon as possible, without interfering existing connections. The detailed LDP state information recovery performs in the background is parallel to the normal LDP operations. At the end, the security aspect of the proposed LDP recovery is analyzed.