94 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 6, NO. 1, FEBRUARY 1998 Cost Analysis of Multicast Transport Architectures in Multiservice Networks K. Ravindran and Ting-Jian Gong Abstract—The paper provides a cost analysis of multicast chan- nels in terms of transport resources allocated by the network. The analysis takes into account the diverse transport requirements of applications in multiservice networks such as multisource broadcasting of data to a common set of destinations, bidirec- tional/unidirectional data transfers among entities, and variable transfer rates of data. The cost model consists of mapping the transport attributes to resource demands and computing the network-wide resource consumptions for data transport. The cost analysis is independent of the speciﬁcs of the backbone network transporting the multicast data and, hence, can provide a network-independent measure of the cost-effectiveness of var- ious multicast architectures. The usefulness of the cost model is illustrated by analyzing multicast data transport costs in “group shared tree” (GST) and “source-speciﬁc tree” (SST) architectures, with both empirical and simulation studies. The cost analysis methodology can be useful in the design and/or evaluation of multiservice data transport architectures. It can also offer a basis for the network provider to implement customer billing functions in a “pay-for-service” type of network management environment envisaged for multiservice networks. Index Terms—Backbone network topology, cost-based routing algorithms, data ﬂow costs, link sharing, multicast architectures, multicast routing, multisource ﬂow speciﬁcation. I. INTRODUCTION M ULTICAST transport architectures provide the network capability for multidestination data delivery, as required by distributed applications (e.g., clients accessing a replicated ﬁle service, users interacting with one another in a teleconfer- ence session) [1]. They employ communication channels that transport data from sources to destinations over a network path implemented by switching nodes and internode links. Typically, a transport architecture embodies the following functions: • point-to-multipoint forwarding of data, whereby the data units from each source are routed along a tree-structured path segment leading toward various destinations (trans- port level replication of data); • multiplexing of data, whereby the data units from multiple sources are multiplexed at various intermediate routing Manuscript received September 7, 1995; revised July 10, 1997; approved by IEEE/ACM TRANSACTIONS ON NETWORKING Editor D. Estrin. This work was supported in part by the Information Institute Partnership program between the U.S. Air Force Rome Laboratory and the City University of New York (CUNY), and in part by the U.S. Air Force Rome Laboratory under Contract F30602-94-C-0241. K. Ravindran was with Kansas State University, Manhattan, KS USA. He is now with the Department of Computer Science, The City College of CUNY, New York, NY 10031 USA (e-mail: ravi@cs-mail.engr.ccny.cuny.edu). T.-J. Gong was with Kansas State University, Manhattan, KS USA. He is now with Clear Communications Corporation, Lincolnshire, IL USA. Publisher Item Identiﬁer S 1063-6692(98)01451-4. nodes to ﬂow toward destinations over common down- stream path segments (transport level path sharing across data ﬂows). In Fig. 1, the data from sources and both ﬂow over the tree rooted at node 4. A transport architecture determines how multicast paths are set up to interconnect sources and destinations in an application and how various path segments are shared across the multisource data ﬂows. The transport resources allocated for multipoint-to- multipoint data exchanges in an application are in the form of node buffers and node-to-node link bandwidths in the backbone network, to realize the data path segments. The extent of resource allocations is based on the ﬂow characteristics of data, such as the average data rates. In addition to ﬂow-related resources, each link in a path segment (i.e., hop) incurs link “connection” management overhead, such as maintaining routing information and state descriptors for data ﬂow. For a routing algorithm that sets up paths, the link overhead and ﬂow-related resource allocation on a hop manifest as distinct cost components. The link overhead is a ﬁxed cost since it does not depend on the number of ﬂows multiplexed along this hop or on the transport characteristics of these ﬂows. In general, if the data of more than one source are allowed to share a path, the cost of combined ﬂows over this path can be less than the sum of the costs of individual ﬂows, due to amortization of ﬁxed cost across the component ﬂows. The cost of a path can be reduced by increasing the number of ﬂows sharing this path. Referring to Fig. 1, the data of and share resources allocated in the tree rooted at node 4, but not resources in the path between nodes 5 and 4. By shifting the root of the shared tree to node 5, the overall cost of data distribution can be reduced. With cost reduction due to path sharing across multiple ﬂows, it is possible that there exist paths with more number of hops but less total cost. 1 Thus multicast transport cost intricately depends on: • ﬂow characteristics of data (such as data rates); • placement of source and destination entities in physical topology of the network; • extent of path sharing achievable in the chosen transport architecture. 1 An example of this analogy is the “ride-share” by commuters traveling to workplaces. Consider two commuters and traveling to a workplace . They may ride in separate cars to a “park-and-ride” lot and travel therefrom in a single car to . In many cases, this car-sharing may be more cost-effective than and riding in separate cars all the way up to , even though the total distance traveled to and then on to may be higher for and/or . The cost-effectiveness stems from amortization of ﬁxed costs such as highway tolls, car wear-and-tear, and driver fatigue. 1063–6692/98$10.00  1998 IEEE