Understanding the bgp transport delay

Border Gateway Protocol (BGP) [35] tied together the Internet's global routing infrastructure. In BGP, routers exchange routing updates to adapt to connectivity changes caused by either intentional routing policy changes or, more commonly, unexpected software and hardware failures. To maintain the continuous reachability, BGP update exchange (or delivery) is expected to be reliable and fast. However, it has long been anecdotal knowledge that such BGP update delivery could be slow, particularly in sending the whole BGP routing tables. Through a longitudinal assessment on BGP data collected by RouteViews and RIPE for over eight years, we show empirical evidence that BGP table transfer can take up to tens of minutes to complete. The observation is prevalent across data collected from different time and Internet locations. Nevertheless, the explanation of these delays remains unclear, and it presents a considerable challenge to identify and understand the causes behind these slow times. The main goal of this dissertation is to gain a better understanding of these delay times from the perspective of BGP transport behavior. Although there have been a plethora of studies on TCP performance in supporting of various applications, relatively little is known about the interaction between TCP and BGP, which is exactly an application running on top of TCP. In this work, we thoroughly investigate the actual BGP/TCP data from operation networks. We find recurring transport problems, including those reported in previous literature and new issues identified in this work that collectively induce significant delay. Such problems are due to various reasons, including implementation issues, router features, and the interaction between BGP and TCP, to name a few. To further quantify and analyze the delay, we develop T-DAT, a tool that can be deployed together with BGP data collectors to infer various factors behind the observed delay, including the BGP's sending and receiving behavior, TCP's parameter settings, TCP's flow and congestion control, and the network path limitation. Applying T-DAT on TCP trace, we reveal and characterize the principal contributing factors behind delay times in the collected BGP dataset. Identifying these delay factors makes a significant step for ISPs and router vendors to diagnose and improve the BGP table transfer performance. Admittedly, given the scale and heterogeneity of the current BGP network, no single study may answer every question of BGP transport delay in the wild. This work on understanding BGP transport delay not only identifies and explains BGP slow delivery problem in actual operation networks, but also sheds lights on the need of improving BGP monitoring practice based on observing detail TCP level dynamics. As an important contribution of this work, we develop a BGP analysis tool suite to augment the current BGP monitoring settings and enable in-depth studies of BOP transport behaviors.