Content & DNS
Internet content: Content is king in today's Internet. Most of the traffic exchanged in the Internet relates to content whose popularity changes very fast over time. Users define which content is « hot », while distributed applications and the content replication platforms that sometimes host them ensure scalable content delivery to end-users. To understand where Internet content is and how it is being distributed, we need to map the content infrastructure, and how different applications rely on the content infrastructure to make their content available. In this work, we rely on active measurements to map the infrastructure hosting content in the Internet. This is joint work with Bernhard Ager, Georgios Smaragdakis and Wolfgang Mühlbauer.
DNS in the Wild: The Domain Name System is nowadays being (mis)used by CDNs to map user resquests to appropriate servers. DNS is therefore the central mechanism through which the location from which content is obtained is being chosen. Despite its widespread use for several decades, the behavior of DNS deployment within ISP networks is still poorly understood. In this work, we measure the performance of DNS from a large set of vantage points inside commercial ISPs. This is joint work with Bernhard Ager, Georgios Smaragdakis and Wolfgang Mühlbauer.
OpenFlow: The granularity at which routing paths are computed today is static. Destination prefixes are used as the main granularity at which packets are handled in the Internet. OpenFlow aims at giving more flexibility in the way traffic is handled, by allowing data forwarding decisions based on any subset of the packet header, or even the content of the packets. Together with our lab in the US, we are investigating several questions related to the granularity at which data forwarding should be done, and where a given forwarding granularity should be used inside the network. This is joint work with Anja Feldmann, Nadi Sarrar, Robert Sherwood, and Xin Huang.
Scalable routers: As virtual routers become more popular, virtualization platforms need to be able to scale their forwarding. We present a method to scale virtual routers based on TCAMs. We rely on a merged data structure that enables the sharing of prefixes from several FIBs in TCAMs. This is joint work with the group of Gaogang Xie, Laurent Mathy and Kave Salamatian.
BGP dynamics: In order to decrease the amount of exchanged routing messages and transient routes, BGP routers rely on MRAI timers (RFC4271) and route flap damping (RFC2439). These timers are intended to limit the exchange of transient routing messages. In practice, these timers have been shown to be partly ineffective at improving convergence, making it even slower in some situations. In this work, we propose to add timers to routing protocols that enforce an ordering of the routing messages such that path exploration is drastically reduced while controlling convergence time. Our timers, called MRPC (metrics and routing policies compliant), are set independently by each router and depend only on the metrics of the routes received by the router as well as the routing policies of the router. No sharing of information about routing policies between neighboring ASs is required by our solution. This is joint work with Marc-Olivier Buob and Anthony Lambert.
Scalable route redistribution inside large ASs: The number of routes to be redistributed inside large transit ASes closes the million. Not only does the storage and propagation of this information poses serious scalability issues, but the way routers select routes will increase in complexity, requiring more sophisticated techniques than the current BGP in order to let each router leverage the routing information available to the AS. This is joint work with Cristel Pelsser, Mickael Meulle, Iuniana Oprescu, and Olaf Maennel.
Routing and traffic interactions between ASs: The behavior of a single AS is relatively well understood. Much less is known about how neighboring interact. In this work, we quantify routing interactions between two particular Ass. This is joint work with Renata Teixeira and Christophe Diot.
Modeling the Internet-wide routing of the Internet: Not much is known today of how BGP routes are propagated across the Internet. Before trying to change the interdomain routing architecture, one should know what aspects are important in the routing design. We thus aim to find out what are the important aspects when trying to reproduce the result of the route filtering and propagation of the BGP routes across the whole Internet. For that purpose, we build models of Internet routing and try to use them to infer general properties of the routing system, e.g., what level of detail is necessary, how optimal the current routes are. This is joint work with are working with Wolfgang Mühlbauer, Olaf Maennel, Anja Feldmann and Mickael Meulle.
Bogon filtering: ISPs often use filters to protect themselves and their customers from non-legitimate prefixes (often called bogon prefixes). Unfortunately, some ISPs still do not configure their filters via the lists published by the RIRs. Instead, they choose to manually configure filters. As allocated address space changes over time, bogons become legitimately announced prefixes. However, not all filters by all ISPs are updated at the appropriate time. The goal of our work is to develop a methodology that detects where wrongly configured filters exists, so that ISPs can be contacted and asked to update their filters. This is joint work with Randy Bush, Olaf Maennel, and Matt Roughan.
Internet optometry: Bogon filtering is one particular reachability problem. When trying to assess the status of reachability of arbitrary subsets of the address space, limitations in terms of visibility of measurements are particularly annoying. BGP for example is biased towards the customer-provider part of the Internet, and publicly available BGP observation points are mostly located in the core of the Internet, not the edge. This means that we have a very partial view of how well reachability works in the Internet. In this work, we try to identify the biases that affect visibility of reachability in the Internet and develop methodologies to counter those biases. This is joint work with Randy Bush, Olaf Maennel, and Matt Roughan.
Topological properties of the AS-level Internet: The current AS-level Internet topology is believed to have properties like power-law in the node degrees, a loose hierarchy due to peering relationships. Part of this topology is unknown, as only limited obervations of BGP paths are available. The true properties of the AS-level Internet might thus never be known. Still, obtaining a better understand of the AS-level graph is important to understand its evolution. In this work we try to understand the properties of the AS-level Internet by studying topology generators and how different the properties of the generated topologies are compared to observed ones. This is joint work with Hamed Haddadi, Damien Fay, Andrew Moore, Richard Mortier, Almerima Jamakovic, and Miguel Rio.
Multi-scale properties of Internet traffic: During the 90's, several scaling behaviors have been identified in Internet traffic, including self-similarity, long-range dependence, and cascades. Because different behaviors occur at different and sometimes overlapping timescales, distinguishing them requires much methodological care. Wavelet-based estimators like the Logscale Diagram and the Multiscale Diagram have been invaluably useful. However, when scaling properties mix with non-stationary behavior, biases may happen and mislead the traffic analysis. In this work, a new methodology that distinguishes between different types of scaling processes has been developed. Thanks to this methodology, different scaling behaviors have been identified in flow arrivals.
Traffic models over the AS-level topology: Most of current traffic models deal with the dynamics on a single link. Understanding the relationship between the control plane and the data plane requires that we have means to simulate how the traffic gets spread over the global Internet. This is clearly a long-term goal to be able to reproduce both the control plane and the data plane. Short-term goals of this project is to first find ways to model how the traffic on the AS topology changes over time.
Traffic engineering with BGP: BGP has not been designed to be used as a mechanism for traffic engineering. However, its flexibility on the selection of routes makes it possible to adjust routing to fit traffic engineering goals. In this work, we have developped traffic engineering techniques for small and large networks. We have also evaluated the effectiveness of different types of traffic control: outbound traffic control is far easier than in the inbound direction. This is joint work with Cristel Pelsser, Bruno Quoitin and Olivier Bonaventure.
Geolocation of Internet hosts: IP Geolocation consists in finding the geographic location of an IP address in the Internet. There are two methods: passive (geolocation databases) and active (active measurements). Geolocation databases are known to have limited accuracy, while active measurements place burden on the network and require distributed measurements infrastructures. Both methods have limited accuracy, but their actual accuracy has been understudied in the literature, especially for geolocation databases. In this work, we investigate the accuracy of geolocation databases based on ground truth information, and find that their accuracy is very limited at the country-level. This is joint work with Ingmar Poese, Bamba Gueye, Benoit Donnet and Mohamed Ali Kaafar.