Monitoring Multicast on a Global Scale: Project Overview, History, Publications and Results

Project Overview      Ongoing Data Collection      Publications      Important Results

Project Overview:

MANTRA (Monitor and Analysis of Traffic in Multicast Routers) is a tool for the global monitoring of the multicast infrastructure at the network layer. MANTRA was developed by Prashant Rajvaidya while pursuing his Ph.D. in the Department of Computer Science at the University of California Santa Barbara and during his internship at the Cooperative Association for Internet Data Analysis (CAIDA) at the San Diego Supercomputing Center (SDSC). MANTRA, as well as the rest of Prashant's Ph.D. work, was supervised by Kevin C. Almeroth and supported by grants from Cisco Systems and the National Science Foundation (NSF).

MANTRA collects data from multiple routers, aggregates collected data to generate global views, and processes these views to generate monitoring results[4]. Data is collected from the routers by logging into the routers and capturing their internal memory tables[5]. The type of data collected from these routers ranges from routing tables for MBGP and DVMRP, to forwarding tables for PIM, to source announcements for MSDP[6].

Results from MANTRA have been used to monitor various aspects of multicast including multicast deployment[3], network topology characteristics[2], MSDP operation[1][4], and data distribution trees[4]. Some of the applications of these results include traffic analysis, route analysis, analysis of host-group behavior, and routing problem debugging.

Ongoing Data Collection:

During the first 5 years of its operation (1998-2003), MANTRA was used to monitor 18 routers including several key exchange points and border routers for transit providers. During this period, the granularity of monitoring results was 15 minutes; i.e. fresh data was collected and results were updated at 15 minute intervals. These results were available in near real-time online in the form of descriptive tables, static plots, interactive topology maps, and interactive geographic visualization.

Since 2003, the scope of MANTRA has been limited to collecting only MBGP information from a few routers in the Internet2 backbone. The goal behind this data collection is to conduct a long-term study of the evolution of multicast. We have, however, discontinued real-time data processing and the real-time results are no longer available.

Project Publications:

Listed below are the publications that have been generated by this work:

Important Results:

Over the lifetime of MANTRA, we have collected over 3 million routing tables from 19 different locations accounting for approximately a terabyte of data. We have analyzed this data to track multicast evolution, estimate usage, monitor topology changes, and troubleshoot routing problems. The following are some of the key findings and results based on our work[2][4]:

The Size of the Active Infrastructure

The figure on the right shows the active address space. We define active address space as the number of addresses that have been announced at least once prior to the measurement period and will be announced at least once after. The following are the key conclusions from these results:

  • Deployment doubled during the 4 year period shown in the graph, to about 40 million addresses.
  • Deployment had been slowly but consistently decreasing since July 2001 (but still twice what it was from when we first started collecting data in 1999).

Session and Membership Statistics
Active multicast address space (Infrastructure Size).

Connectedness of the Infrastructure

The figure on the right shows the connected address space. We define connected address space as the number of addresses represented by valid route announcements and, therefore, connected to the MBGP topology during each measurement interval. While active address space (above) represents all the addresses that can possibly be connected, connected address space represents all the addresses that are actually reachable during the measurement instance. The following are the key conclusions from these results:

  • Significant Increase in Infrastructure Connectivity: Connectivity was poor between 1999 and 2001 because only a fraction of active address space was connected at any one time. Towards the end of the measurement period, however, connectivity increased significantly with over 95% of the the active address space connected to the MBGP topology at any given instance.
  • Routing Stability Problems: Routing was not particularly stable during much of the 4 year period because connectedness within the infrastructure varied to a high degree. This instability created significant problems for application developers and network administrators. As a result, application developers and network providers tried multicast but did continue to use it.

Session and Membership Statistics
Address space connected to the Infrastructure.

Growth in Native Multicast Deployment

The figure on the right shows the growth of the multicast infrastructure. This graph shows a line that increases each time a new address or group of addresses is first announced via an MBGP route announcement. Growth results show that there is a clear rise in the size of the infrastructure. The following are some of the key observations and conclusions:

  • The amount of address space has grown by nearly 50 times over the four year period from 50 million to about 2.5 billion addresses.
  • Up until March 2000, growth has been in spurts.
  • Only gradual growth has occurred since then.
  • Only small networks have joined the topology recently.

Session and Membership Statistics
Growth of the MBGP-reachable address space.

Loss in Native Multicast Deployment

The figure on the right shows the loss of networks in the infrastructure. That is, networks that stopped announcing their MBGP routes. This graph takes as its starting point the total number of addresses announced via MBGP over the four year measurement period. The graph then decrements this number when an MBGP advertisement is observed for the last time. The basic idea is: "growth shows the cumulative number and timing of newly advertised address space while loss shows when address space stops being advertised". The following are some key observations and/or conclusions:

  • There was significant loss (92%) during the 4 year period.
  • Most of the networks lost were the unstable and transient. This loss occurred because system administrators could not maintain an acceptable level of service, multicast was essentially turned off.
  • The most significant drop occurred in March 2000 during the dismanteling of the MBone. That is, as use of the DVMRP-based MBone declined, fewer service providers were willing to offer tunnels. And because many service providers did not offer native multicast, many (public) Internet users had no way of receiving multicast content.

Session and Membership Statistics
Loss of the MBGP-reachable address space.

The End of MBone

As the tunnel-based MBone was replaced with native multicast, the MBone essentially ceased to exist. The figure on the right shows the loss in the address space reachable via the DVMRP-based MBone at FIXW between November 1998 and July 2000. The graph in this figure takes as its starting point the total number of unique addresses that were advertised via DVMRP during the collection period. The graph then decrements at the point in time when each address stops being advertised. This figure confirms that the use of DVMRP started to decline in February of 1999 and almost completely ceased by March 2000.

Session and Membership Statistics
Loss of the DVMRP-reachable address space.