Configurable active multicast congestion control☆
Introduction
Multicast is an efficient way to support group-based applications. A multicast congestion control scheme is indispensable for a multicast flow to efficiently and “fairly” compete for network resources with unicast flows and other multicast flows. Multicast congestion control is a hard problem. There are two main challenges. Given the ubiquitous presence of TCP traffic in the Internet, the first challenge is how to compete for the bandwidth with TCP flows. This challenge in essence is how to define and implement “fairness” between multicast flows and unicast flows. Previous research [7], [9], [10], [21], [26] discusses the fairness from different points of view. Though each of them gives an insight to the essence of a congestion control algorithm, a formal congestion control algorithm to meet the exact fairness criterion is hard to realize. Most of the proposed congestion control schemes [1], [2], [4], [5], [19], [20], [25] try to conform to the practical criterion “TCP-friendliness [17]”, which stipulates that the arrival rate of a non-TCP flow should not exceed that of a conformant TCP flow in the same circumstances. Given that a multicast session may involve a large number of receivers which may experience different degrees of congestion due to dynamic network traffic and heterogeneous link capacities, the second challenge is how to satisfy the specific interest of heterogeneous receivers in the same multicast session. This challenge is essentially how to define the “fairness” among receivers. Refs. [15], [16] give a definition of inter-receiver fairness. The generic approach to address inter-receiver fairness is partitioning receivers based on their available bandwidth into multiple groups each of which has a different rate [5], [6], [8], [27]. With a pure end-to-end approach, it is hard to address both “TCP-friendliness” and “inter-receiver fairness” for a multicast communication.
The “end-to-end argument [12]” is a main design principle for existing unicast and multicast congestion control schemes. For the congestion control case, it stipulates that the congestion detection and avoidance mechanism should be conducted only by end nodes. This design principle has advantages to insulate applications from the complexity of the underlying communication networks. However, it has the limitation that it prevents applications from exploiting detailed knowledge of the underlying network for performance enhancement. Previous research [13] shows that it is compatible with the end-to-end argument to provide generic customizable service interfaces to users in the intermediate network. More and more researchers agree that a better congestion control scheme can be designed with the help of routers. The trend started initially by designing better queueing mechanisms such as RED [49] and REM [50] and followed by more complex unicast congestion control schemes such as ACC [45], [46] and XCP [48]. RMANP [43], [44] and AMCA [47] are the “active” version of a multicast congestion control scheme. Following the same design philosophy of RMANP and AMCA, this research is based on the similar assumption that the future routers can provide some functionalities and resources for application protocols. These routers are called “active routers” in this paper. They could provide a full fledged active platform as dictated by active networks [14] or they could be an enhanced router with limited system support to some application protocols such as multicast.
In this paper, we present a novel active multicast congestion control scheme (AMCC), which can be used by both reliable multicast applications and unreliable multicast applications. It is designed to achieve good performance with support from active routers. It can adapt to different needs of different multicast applications with simple configuration. AMCC has two main contributions: (i) AMCC can behave both as single-rate multicast and multi-rate multicast. AMCC is the first congestion control scheme to achieve TCP-friendliness and inter-receiver fairness using only one multicast group. With flexible configuration, AMCC can operate at one of three modes: single-rate congestion control, per-node congestion control and per-link congestion control. It is thus able to adapt to different application needs. (ii) AMCC can be used by both reliable and unreliable multicast applications. In particular, when it is used by reliable multicast applications, AMCC has a special mechanism to regulate repair packets which is not specifically addressed by previous congestion control schemes. AMCC can also achieve fair competition among a number of multicast flows when each of them requires the same amount of resources.
The rest of the paper is organized as follows. In Section 2, we summarize existing related work. In Section 3, we present the protocol details of AMCC. In Section 4, we describe our simulation results. In Section 5, we discuss some protocol features. The paper concludes in Section 6.
Section snippets
Related work
Fairness must be addressed by any congestion control scheme. There are two main theoretical definitions of fairness to dictate bandwidth sharing among multiple network flows: max–min fairness [24] and proportional fairness [10]. It is difficult to construct an effective algorithm to realize the theoretical fairness criteria. Based on the fact that a multicast flow may involve a large number of receivers and potentially could consume significantly fewer resources per receiver than unicast flows
Detailed description of active multicast congestion control
In this section, we elaborate on the details of our protocol and discuss the design alternatives for particular mechanisms. The basic idea of active multicast congestion control is that the source or active routers dynamically choose the representative receiver from all their downstream receivers and use the feedback from the fastest or slowest receiver to adjust forwarding rate to adapt to different degrees of congestion. AMCC can act as a single-rate multicast congestion control scheme by
Simulation results
We use simulation to evaluate the performance. We implement AMCC in NS2 [41] and use the PANAMA active package [42] to simulate the function of active routers.
We have done a wide range of simulation experiments to investigate the behavior of AMCC under different network conditions. In all the experiments, we are interested in how the throughput or average throughput of a node evolves with time and we are concerned with the bandwidth sharing among different multicast and unicast flows in both
Processing cost at active routers
The benefits of AMCC come at a cost. Specifically, active routers need more storage and computing power to store states and perform congestion control operations. The storage spaces decide the capability of an active router to smooth the congestion. Different applications have different requirements on congestion smoothing capability of active routers. Therefore, the exact amount of storage space at an active router actually depends on the requirements of multicast applications. From Eq. (4),
Conclusions and future work
In this paper, we have presented a novel active congestion control scheme for multicast applications. Our protocol can be flexibly configured to perform three types of congestion control: single-rate, per-node and per-link. It can achieve TCP-friendliness and inter-receiver fairness under different conditions using only one multicast group, which is starkly different from previous multi-rate schemes which need to use multiple multicast groups to achieve inter-receiver fairness. Our congestion
Acknowledgements
The authors would like to thank network research group in University of Massachusetts, Amherst for their PANAMA software package. We are also grateful to anonymous reviewers for their constructive suggestions and useful pointer to references to ACC [45], [46], RMANP [43], [44] and AMCA [47].
Baochun Bai holds a BEng from Huazhong University of Science and Technology and an MEng from Chinese Academy of Sciences and an MSc from University of Alberta. He is now a Ph.D. student in University of Alberta. His current research interest is computer networks, multimedia systems and video compression.
References (50)
- et al.
Scalable fair reliable multicast using active services
IEEE Network Magazine
(2000) - L. Rhee, N. Balaguru, G.N. Rouskas, MTCP: scalable TCP-like congestion control for reliable multicast, in: Proceedings...
- S. Bhattacharyya, D. Towsley, J. Kurose, The loss path multiplicity problem for multicast congestion control, in:...
- S. Shi, M. Waldvogel, A rate-based end-to-end multicast congestion control protocol, in: Proceedings of ISCC...
- L. Vicisano, L. Rizzo, J. Crowcroft, TCP-like congestion control for layered multicast data transfer, in: Proceedings...
- S. McCanne, V. Jacobson, Receiver-driven layered multicast, in: Proceedings of ACM...
- D. Rubenstein, J. Kurose, D. Towsley, The impact of multicast layering on network fairness, in: Proceedings of ACM...
- S. Bhattacharyya, J. Kurose, D. Towsley, R. Nagarajan, Efficient rate-controlled bulk data transfer using multiple...
- M. Vojnovic, J.Y. Le Boudec, C. Boutremans, Global fairness of additive-increase and multiplicative-decrease with...
- et al.
Rate control in communication networks: shadow prices, proportional fairness and stability
Journal of the Operational Research Society
(1998)
End-to-end arguments in system design
ACM Transactions on Computing Systems
A survey of programmable networks
ACM SIGCOMM Computer Communication Review
Promoting the use of end-to-end congestion control in the Internet
IEEE/ACM Transactions on Networking
The performance of TCP/IP for networks with high bandwidth-delay products and random loss
IEEE/ACM Transactions on Networking
Data Networks
Cited by (4)
Layered multicast congestion control by particle swarm optimization in heterogeneous environment
2011, Journal of China Universities of Posts and TelecommunicationsRobust image watermarking using DWT and SVD for copyright protection
2017, AIP Conference ProceedingsWatermarking technique for wireless multimedia sensor networks: A state of the art
2012, ACM International Conference Proceeding SeriesALMTF++: A new congestion control for large scale multicast transmission
2009, WebMedia 2009 - 15th Brazilian Symposium on Multimedia and the Web
Baochun Bai holds a BEng from Huazhong University of Science and Technology and an MEng from Chinese Academy of Sciences and an MSc from University of Alberta. He is now a Ph.D. student in University of Alberta. His current research interest is computer networks, multimedia systems and video compression.
Janelle Harms is an Associate Professor in the Computing Science Department of the University of Alberta. She received her Ph.D. in Computer Science from the University of Waterloo, Canada in 1992 working in the area of resource allocation and performance analysis of networks. Her research interests include performance aspects of network resource allocation, routing and network design problems in wireless sensor networks, mobile ad hoc networks and fixed-topology networks.
Yuxi Li is a postdoctoral fellow with the Alberta Ingenuity Centre for Machine Learning (AICML), in the Computing Science Department of the University of Alberta. He is investigating machine learning techniques for problems in quantitative finance. He received his Ph.D. in the Computing Science Department of the University of Alberta, Canada in 2006. His thesis was on network performance optimization with changing and uncertain traffic demands, for both the internet and wireless networks. He is interested in performance optimization for communication networks. During his Ph.D. study, he received the prestigious Honorary Izaak Walton Killam Memorial Scholarship, ARC Karl A Clark Memorial Scholarship, and Queen Elizabeth II Graduate Scholarship.