Multicast performance measurement on a high-performance IP backbone☆
Introduction
The very-high-speed Backbone Network Service (vBNS) is a high-performance research and education network sponsored by the US National Science Foundation (NSF) through a cooperative agreement with WorldCom. The network was chartered in 1995 to serve as a proving ground for new technologies, to support the development of advanced services, and to support the research conducted by its user community. Consistent with this mission the vBNS network has continually evolved over the past five years and has made significant contributions in the areas of IP multicast, traffic measurement and analysis, IP version 6, multi-protocol label switching, and IP quality of service. Two of the most noteworthy areas of concentration throughout the evolution of the vBNS are traffic measurement and IP multicast deployment. This paper describes recent work that marries these two areas with the creation of a unique IP multicast performance measurement capability on the vBNS backbone. In keeping with previous measurement work on the vBNS, performance results are calculated entirely on an end-to-end basis, without relying on data from the network elements under test.
This paper presents the design and implementation of a multicast active measurement system, as well as results of measurements taken on the vBNS backbone. Section 2 provides an overview of the vBNS network to provide context for the work. Section 3 relates the history of IP multicast deployment on the vBNS, concluding with a description of the current multicast configuration. Section 4 describes the existing active and passive measurement capabilities in place on the vBNS backbone. Section 5 recounts the development and implementation of an active multicast measurement system that is capable of both generating synthetic multicast traffic at varying rates and measuring packet loss and reordering at each of a number of receivers. The paper concludes with a set of representative results from the measurement system running on the native-multicast-enabled vBNS backbone.
Section snippets
vBNS overview
The vBNS is a high performance IP backbone that spans the continental United States. The network provides high speed interconnection among supercomputer centers, universities, research and education institutions, and other research networks in the US and abroad. Sites are connected to the vBNS backbone at speeds ranging from 45 Mbps (Megabits per second) (DS-3) up to 622 Mbps (OC-12c) and most typically at 155 Mbps (OC-3c).
There are 13 vBNS points of presence (POPs) in cities across the US that
vBNS IP multicast evolution and architecture
The vBNS has offered a multicast service since its inception in 1995. Originally this offering consisted of Distance Vector Multicast Routing Protocol (DVMRP) tunnels between vBNS-owned DEC Alpha hosts located at five vBNS-connected supercomputer sites. In 1996 the DVMRP tunnels to customer sites were re-homed to vBNS Cisco 7500 series routers. In turn, these Cisco routers, which at the time were the vBNS core routers, began running native IP multicast internally via Protocol Independent
vBNS measurement capabilities
One of the distinguishing features of the vBNS network is its extensive measurement infrastructure. The vBNS backbone is highly instrumented to facilitate both active and passive traffic measurements. The Sun hosts located in each vBNS POP support active measurements of both throughput and round-trip times. Automated scripts conduct throughput measurements nightly on both directions of each vBNS backbone trunk. The throughput performance tests consist of both congestion-controlled TCP traffic
Multicast measurement
The multicast measurement system described here is, in a sense, an extension of the active, unicast performance measurements described above. The goal was simply to extend the performance measurement capability into the multicast realm to provide statistics on the performance of the multicast backbone. The same Sun hosts used for the round-trip time and unicast throughput measurements serve as the platform for the multicast measurements. The tests are implemented via custom developed software
Test results
In this section, we first give details of the tests whose results are reported in this paper. Second, we present the output that a single test suite produces and examine the loss pattern of two different receivers. Next, we provide an analysis of the effects of the transient phase during which multicast state is installed. Finally, we examine the correlation between the loss rates receivers experience and their relative depth in the multicast distribution tree.
Related work
Measurement and monitoring of Internet multicast infrastructure is an active area of research. One closely related activity is Multicast Inference of Network Characteristics (MINC) [4]. Using end-to-end multicast probes, synthetic multicast traffic is introduced and measured with end-to-end losses measured by the receivers and correlated to infer loss behavior of the links in the multicast distribution trees. Estimators for link loss rates have been developed, and a network visualization tool
Conclusions and future work
The automated multicast tests described in this paper provides the vBNS with historical data on how the network is performing as well as revealing performance or configuration problems proactively. The tests are a valuable debugging tool when used interactively.
Many methods of correlation were attempted to approximate the nature of the loss. The most successful approach involved removing the initial loss, attributable to the join latency and shared tree to shortest-path-tree switchover, and
Acknowledgements
The authors would like to acknowledge the insightful comments and suggestions of many people, especially Bill Fenner, Vint Cerf, Ashley Shi, Tom Pusateri, Allison Fisher, Bill Kroah, Kevin Almeroth and Zaid Albanna.
Robert Beverly is a senior engineer with WorldCom's Advanced Internet Technology group in Ashburn, Virginia, where he has worked for the past two years on the very-high-performance Backbone Network Service (vBNS). His primary interests include network performance, statistics and visualization. He received his BS in computer engineering from the Georgia Institute of Technology in 1996.
References (9)
The evolution of multicast: from the MBone to inter-domain multicast to Internet2 deployment
IEEE Network
(2000)- K. Almeroth, L. Wei, D. Farinacci, Multicast reachability monitor (MRM), Internet Engineering Task Force Internet...
- J. Apisdorf, K. Claffy, K. Thompson, R. Wilder, OC3MON: flexible, affordable, high-performance statistics collection,...
- et al.
Multicast-based inference of network-internal loss characteristics
IEEE Transactions on Information Theory
(1999)
Cited by (2)
Evaluating IPv6 on a large-scale network
2006, Computer CommunicationsCitation Excerpt :Performance measurement is valuable because it provides historical data on how the network is fulfilling its objectives as well as revealing performance or configuration problems proactively [11].
A IPv6 network performance test system using multi-agent
2007, 2007 8th International Conference on Electronic Measurement and Instruments, ICEMI
Robert Beverly is a senior engineer with WorldCom's Advanced Internet Technology group in Ashburn, Virginia, where he has worked for the past two years on the very-high-performance Backbone Network Service (vBNS). His primary interests include network performance, statistics and visualization. He received his BS in computer engineering from the Georgia Institute of Technology in 1996.
Gregory J. Miller is the Director of Advanced Internet Technology at WorldCom in Ashburn, Virginia. He leads a team that engineers and operates WorldCom's very-high-performance Backbone Network Service (vBNS). The group is engaged in the development of advanced IP technologies including IPv6, IP multicast, and IP QoS. Prior to Joining MCI (which was acquired by WorldCom), he was a Senior Member of the Technical Staff in the Networking Center at the MITRE Corporation. During his three years at MITRE, he worked on a variety of networking projects, including the development of TCP enhancements for the space environment. He received a BS in Computer Science from Loyola College in Maryland in 1988, and the MS and PhD degrees from the University of Delaware in 1990 and 1993, respectively, both in Computer and Information Sciences.
Kevin Thompson is a Senior Manager for Statistics and Measurement in the Advanced Internet Technology Group at WorldCom in Ashburn, Virginia. Prior to joining MCI (which was acquired by WorldCom), he was a Member of the Technical Staff in the Networking Center at the MITRE Corporation. He received a BS in Computer Science from the University of Virginia in 1987 and a MS in Computer Science from the George Washington University in 1992.
- ☆
This work was supported by the National Science Foundation under Cooperative Agreement Number NCR-9321047.