next up previous
Next: Methodology Up: Visualizing the Global Topology Previous: Visualizing the Global Topology

Introduction

The Internet is evolving to support a variety of distribution models and carry a much broader range of traffic than was traditionally available, in particular multimedia services such as real time audio and video. Multicasting, a network distribution methodology which efficiently transmits data to multiple receivers, is now being widely deployed to solve some of the problems inherent in these new services [5]. Some current uses of the MBone include distributing audio and video streams of conferences, seminars, interactive meetings, and media events. It may see increasing usage as a communication mechanism for networked multi-user 3D applications.

Production Internet routers are only recently beginning to support native multicast. The MBone was designed to provide interim support by using virtual logical links between multicast-capable routers that cross over non-multicast capable infrastructure. Endpoints of these virtual links, or tunnels, encapsulate multicast packets into normal IP packets and route them to other multicast capable subnets over the unicast path between the tunnel endpoints. Tunnel endpoints are typically Unix workstations with IP multicast support, using the multicast routing daemon mrouted. mrouted runs a multicast routing protocol that dynamically manages a spanning tree over the links in the MBone. This tree ensures that data is sent only once across each multicast link, replicated only as necessary when the spanning tree splits into multiple paths.

The MBone, like the Internet itself, grew without planning from a central authority, and shares many of the strengths and weaknesses of the larger Internet. Setting up a tunnel requires minimal coordination, and absolutely no approval by or even informing of other network operators or users, including one's own Internet service provider. Establishing a tunnel requires only that two sites already connected via the Internet exchange IP addresses and configure their machines as endpoints for a tunnel between them. This flexibility has the advantage of facilitating ubiquitous participation without third-party authorization. However, it has also inadvertently spawned an increasingly convoluted and redundant topology, where several copies of the same multicast traffic sometimes travel over the same physical link.

As the size of the MBone grows and Internet congestion becomes increasingly problematic, there is greater incentive to tune tunnel placement to provide the most efficient distribution topology and minimize the imposed Internet workload. Ideally, tunnels should be placed so that they travel along uncongested links that carry no other tunnels. The rapidly changing nature of Internet complicates this task. Two hosts that were connected via an uncongested high bandwidth path six months ago may easily be traveling a different more highly congested route today.

Tunnel placement should match the structure of the existing unicast Internet topology whenever possible, and only traverse non-multicast capable infrastructure long enough to get to the nearest available multicast capable subnet. In practice, however, tunnels have been used to create an overlay network that in some places is disjoint from the underlying Internet topology. The multicast routing protocol determines the shortest paths in its spanning tree based on the tunnel hop count rather than the underlying unicast hop count, resulting in sub-optimal distribution of multicast data.

Given the increasing popularity of the MBone among Internet users and the potential impact of tunnel traffic on the production Internet, Internet service providers (ISPs) are beginning to recognize issues associated with large scale multicast distribution. Providers trying to offer multicast services or engineer transit tunnel traffic currently have very few tools with which to analyze the overall system or the role of their own infrastructure within it.

This inability to effectively understand and manage the MBone, whether by the users, providers, or multicast researchers, is the underlying motivation for our desire to visualize its global structure.



next up previous
Next: Methodology Up: Visualizing the Global Topology Previous: Visualizing the Global Topology



Tamara Munzner