Characterizing the performance and stability issues of the AS path prepending method: Taxonomy, measurement study and analysis

Abstract

We consider the autonomous system (AS) path prepending (ASPP) approach to perform load-balancing on inbound traffic for multihomed ASes. Although the ASPP approach has been practiced by many AS operators for a long time, it is surprising that there still lacks a systematic study of the approach and understanding of its effects and performance implications. The purpose of this paper is to provide such an analysis, which serves as a basis to further improve the approach and to understand its limitations. There are two parts to this study. The first one is based on measurement and analysis of the Route Views data, which provides the fact of the “prevalence” of using ASPP approach in the current Internet and its possible impact on Internet routes. In the second part we propose a model of how ASes perform traffic engineering and we also present some fundamental issues of decentralized traffic engineering in the Internet. In our model, each AS performs inbound load balancing to optimize its traffic engineering goals. Some important questions that we address are how these local actions affect the global network performance and whether these ASPP actions interfere with each other and induce instability. We provide some examples to illustrate the performance and stability issues and suggest some guidelines that will help to avoid route oscillations.