An efficient critical protection scheme for intra-domain routing using link characteristics

In recent years, there are substantial demands to reduce packet loss on the Internet. Among the proposed schemes, finding backup paths in advance is considered to be an effective method to reduce the reaction time. Very commonly, a backup path is chosen to be the most disjoint path from the primary path, or on the network level, a backup path is computed for each link (e.g., IPFRR). The validity of this straightforward choice is based on two things. The first thing is all the links may have the equal likelihood to fail; the second thing is, facing the high protection requirement today, it just looks weird to have links not protected or to share links between the primary and backup paths. Nevertheless, many studies have confirmed that the individual vulnerability of the links on the Internet is far from being equal. In addition, we have observed that full protection schemes (In this paper, full protection schemes means schemes (1) in which backup path is a most disjoint path from the primary path; or (2) which compute backup path for each link.) may introduce high cost (e.g., computation). In this paper, we argue that such approaches may not be cost-efficient and therefore propose a novel critical protection scheme based on link failure characteristics. Firstly, we analyze the link failure characteristics based on real world traces of CERNET2 (China Education and Research NETwork 2). The analysis results clearly show that the failure probabilities of the links in CERNET2 backbone are heavy-tailed, i.e., a small set of links causing most of the failures. Based on this observation, we find out two key parameters which strongly impact link criticality and propose a critical protection scheme for both single link failure situation and multi-link failure situation. We carefully analyze the implementation details and overhead for backup path schemes of the Internet today; the problem is formulated as an optimization problem to guarantee the routing performance and minimize the backup cost. This cost is special as it involves computational overhead. Based on this, we propose a novel Critical Protection Algorithm which is fast itself for both the single link failure and the multi-link failure versions. A comprehensive set of evaluations with randomly generated topologies, real world topologies and the real traces from CERNET2, shows that our scheme gains significant achievement over full protection in both single link failure situation and multi-link failure situation. It costs only about 30-60% of the full protection cost when the network relative availability increment is 90% of the full protection scheme.