A server-to-server view of the internet

Computer Communication Review, 1997

We discuss findings from a large-scale study of Internet packet dynamics conducted by tracing 20,000 TCP bulk transfers between 35 Internet sites. Because we traced each 100 Kbyte transfer at both the sender and the receiver, the measurements allow us to distinguish between the end-to-end behaviors due to the different directions of the Internet paths, which often exhibit asymmetries. We characterize the prevalence of unusual network events such as out-of-order delivery and packet corruption; discuss a robust receiver-based algorithm for estimating "bottleneck bandwidth" that addresses deficiencies discovered in techniques based on "packet pair"; investigate patterns of packet loss, finding that loss events are not well-modeled as independent and, furthermore, that the distribution of the duration of loss events exhibits infinite variance; and analyze variations in packet transit delays as indicators of congestion periods, finding that congestion periods also span a wide range of time scales. In x 3 we characterize unusual network behavior: out-of-order delivery, replication, and packet corruption. Then in x 4 we discuss a robust algorithm for estimating the "bottleneck" bandwidth that limits a connection's maximum rate. This estimation is crucial for subsequent analysis because knowing the bottleneck rate lets us determine when the closely-spaced TCP data packets used for our network probes are correlated with each other. (We note that the stream of ack packets returned by the TCP data receiver in general is not correlated, due to the small size and larger spacing of the acks.) Once we can determine which probes were correlated and which not, we then can turn to analysis of end-to-end Internet packet loss (x 5) and delay (x 6). In x 7 we briefly summarize our findings, a number of which challenge commonly-held assumptions about network behavior.

13TH IEEE International Conference on Network Protocols (ICNP'05), 2005

This paper focuses on understanding the scale and the distribution of "state overhead" (briefly load) that is incurred on the routers by various value-added network services, e.g., IP multicast and IP traceback. This understanding is essential to developing appropriate mechanisms and provisioning resources so that the Internet can support such value-added services in an efficient and scalable manner. We mainly consider the number of end-to-end paths or trees intersecting at a router to represent the amount of state overhead at that router. Hence, we analyze the router-level intersection characteristics of end-to-end Internet paths or trees to approximate the state overhead distribution in the Internet. For the reliability of our analysis, a representative, end-to-end router-level Internet map is essential. Although several maps are available, they are at best insufficient for our analysis. Therefore, in the first part of our work, we exert a measurement study to obtain a large size end-to-end router-level map conforming to our constraints. In the second part, we conduct various experiments using our map and shed some light on the scale and distribution of state overhead of value-added Internet services in both unicast and multicast environments.

2010 INFOCOM IEEE Conference on Computer Communications Workshops, 2010

The diversity of end-to-end (e2e) Internet routes has been studied for over a decade, dating back to Paxson's seminal work from 1995. This paper presents a measurement study of this issue and systematically evaluate the diversity of the Internet routes, while revisiting some of the conclusions previously made. Two large scale experiments are used for evaluation, one executed in late 2006 and the second in early 2009, both employ a set of more than 100 broadly distributed vantage points, actively measuring between each other.

Routing policies used in the Internet can be restrictive, limiting communication between source-destination pairs to one path, when often better alternatives exist. To avoid route flapping, recovery mechanisms may be dampened, making adaptation slow. Unstructured overlays have been proposed to mitigate the issues of path and performance failures in the Internet by routing through an indirect-path via overlay peer(s). Choosing alternate-paths in overlay networks is a challenging issue. Ensuring both availability and performance guarantees on alternate paths requires aggressive monitoring of all overlay paths using active probing; this limits scalability. An alternate technique to select an overlay-path is to bias its selection based on physical disjointness criteria to bypass the failure on the primary-path. Recently, several techniques have emerged which can optimize the selection of a disjoint-path without incurring the high costs associated with probing paths. In this paper, we sh...

IEEE Transactions on Dependable and Secure Computing, 2006

Several research studies have been devoted to improving the reliability and performance of the Internet by utilizing redundant communication paths between end points. Multihoming, coupled with intelligent route control, and overlay networks are two main streams in this area of research which attempt to leverage redundant connections of the Internet for increased reliability and performance. However, the effectiveness of these approaches depends on the natural diversity of redundant paths between two endhosts in terms of physical links, routing infrastructure, administrative control, and geographical distribution. Even in the case of redundant paths, if traffic between two hosts is not actually routed along completely disjoint paths, congestion or failure of a single shared link or router can adversely affect the end-to-end performance or availability of all paths. This paper presents an experimental study of path diversity on the Internet, focusing on the impact of path diversity on multihomed and overlay networks. We base our analysis on traceroutes and routing table data collected from several vantage points in the Internet including: looking glasses at 10 major Internet Service Providers (ISPs), RouteViews servers from 20 ISPs, and more than 50 PlanetLab nodes globally distributed across the Internet. Using this data, we quantify the extent of path diversity in multihoming and overlay networks, highlighting the limitations, and also identifying the source of the limitations in these architectures. From the analysis, we learn that both multihoming route control and current overlay networks are not able to ensure path diversity, which makes it very difficult to provide high-availability services even with the use of these systems. We believe that this work provides the insight into building future systems based on understanding path diversity.

2008

RTT has been widely used as a metric for peer/server selection. However, many applications involving closest peer/server selection such as streaming, tree-based multicast services and other UDP and TCP based services would benefit more from knowing one-way delay (OWD) rather than RTT. In fact, RTT is frequently used as as an approximate solution to infer forward and reverse delays by many protocols and applications which assume forward and reverse delay to be equal to half of RTT. In this paper, we compare and contrast one-way delays and corresponding RTTs using a wide selection of routes in the Internet. We first measure the extent and severeness of asymmetry in forward and reverse OWD in the Internet. We then attempt to isolate the causes of OWD asymmetry by correlating OWD asymmetry with the route asymmetry. Finally, we investigate the dynamics of delay asymmetry. We find there exists a weak correlation between the fluctuation of RTT and OWD but a strong correlation between OWD change and the corresponding route change.

2014

Several works over the past few years have shown that the Internet AS-level topology is partially hidden from the current Internet measurement infrastructures. Most have focused on the incompleteness of the connectivity extracted from BGP data. A few have analysed the connectivity collected by traceroute measurement infrastructures showing the amount of connections introduced by traceroute campaigns. None, however, have investigated in detail the underlying rationale, i.e. the economic nature of the Internet. In this paper we fill this gap by analysing five traceroute infrastructures, found to be active in October 2013, with the p2c-distance metric, which is specifically designed to capture the complex economic dynamics that rule the Internet. We found that the traceroute infrastructures that currently run topology discovery measurements (Ark, DIMES and Portolan), together with BGP route collectors, are able to reveal the full connectivity of 23.50% of the Internet core ASes. This is a considerable improvement given that the BGP infrastructure alone is able to cover only 15.90% of the Internet core. This percentage could be increased up to 48.48% if the remaining two infrastructures (Dasu/Ono and RIPE Atlas) performed topology discovery campaigns. We also found that the placement of traceroute probes is not optimal from a topology discovery perspective, as it causes several probes to provide only redundant connectivity information. We show that the same number of traceroute probes optimally deployed, would be able to completely reveal the full AS connectivity of the Internet core.

Mathematical Problems in Engineering, 2017

Internet Protocol (IP) is used to identify and locate computers on the Internet. Currently, IPv4 still routes most Internet traffic. However, with the exhausting of IPv4 addresses, the transition to IPv6 is imminent, because, as the successor of IPv4, IPv6 can provide a larger available address space. Existing studies have addressed the notion that IPv6-centric next generation networks are widely deployed and applied. In order to gain a deep understanding of IPv6, this paper revisits several critical IPv6 performance metrics. Our extensive measurement shows that packet delay and loss rate of IPv6 are similar to IPv4 when the AS-level paths are roughly the same. Specifically, when the link utilization exceeds a threshold, for example, 0.83 in our study, variation of packet delay presents a similar pattern with the variation of link utilization. If packet delay of a path is large, packet-loss rate of that path is more likely to fluctuate. In addition, we conduct a first-ever analysis ...

IEEE/ACM Transactions on Networking, 2008

Despite the architectural separation between intradomain and interdomain routing in the Internet, intradomain protocols do influence the path-selection process in the Border Gateway Protocol (BGP). When choosing between multiple equally-good BGP routes, a router selects the one with the closest egress point, based on the intradomain path cost. Under such hot-potato routing, an intradomain event can trigger BGP routing changes. To characterize the influence of hot-potato routing, we propose a technique for associating BGP routing changes with events visible in the intradomain protocol, and apply our algorithm to a tier-1 ISP backbone network. We show that (i) BGP updates can lag 60 seconds or more behind the intradomain event, (ii) the number of BGP path changes triggered by hot-potato routing has a nearly uniform distribution across destination prefixes, and (iii) the fraction of BGP messages triggered by intradomain changes varies significantly across time and router locations. We show that hot-potato routing changes lead to longer delays in forwarding-plane convergence, shifts in the flow of traffic to neighboring domains, extra externally-visible BGP update messages, and inaccuracies in Internet performance measurements. 1 To obtain up-to-date information about the size of routing tables, please reference http://www.cidr-report.org.

arXiv (Cornell University), 2009

By focusing on what can be observed by running traceroute-like measurements at a high frequency from a single monitor to a fixed destination set, we show that the observed view of the topology is constantly evolving at a pace much higher than expected. Repeated measurements discover new IP addresses at a constant rate, for long period of times (up to several months). In order to provide explanations, we study this phenomenon both at the IP, and at the Autonomous System levels. We show that this renewal of IP addresses is partially caused by a BGP routing dynamics, altering paths between existing ASes. Furthermore, we conjecture that an intra AS routing dynamics is another cause of this phenomenon.