Cost-based Pricing for Multicast Streaming Services

Transcription

1 Cost-based Pricing for Multicast Streaming Services Eiji TAKAHASHI, Takaaki OHARA, Takumi MIYOSHI,, and Yoshiaki TANAKA Global Information and Telecommunication Institute, Waseda Unviersity 29-7 Bldg., Nishi-Waseda, Shinjuku-ku, Tokyo, JAPAN Tel: Fax: Graduate School of Science and Engineering, Waseda University Faculty of Systems Engineering, Shibaura Institute of Technology Abstract: In multicast-type services, many receivers share a single data flow. The question is how to split the cost of such a flow among the receivers. There is a need to provide receivers with reasonable feedback on the cost that their network usage incurs. Therefore, cost sharing methods will play an important role in multicast-type services in the future. This paper discusses cost sharing methods where many receivers share multicast flows and describes how these costs are shared among the members of multicast groups. We consider the case where an information provider (sender) provides receivers with multicast-type video streaming services by using the network resources owned by a telecommunication carrier. First, we consider some cost sharing methods to determine the charge in multicast-type streaming services and then calculate the shared cost to analyze the cost structure of multicast-type streaming services. Key Words: Multicast, Streaming Service, Cost, Pricing I. Introduction Multicast technology is expected to become widely used, since its link sharing capability makes it well suited to broadband multimedia services. In multicast technology, a point-to-multipoint (multicast) connection is used to copy packets only at branching nodes, which ensures network efficiency. However, multicast services have not become popular. One of the main reasons for the unpopularity of multicast is the lack of a good business model and effective pricing mechanism, as the cost structure of network usage is quite different from that of unicast services [1]. In multicast-type services, many receivers share a single data flow. Thus, one of the fundamental questions is how to split the cost of this flow among receivers. There is a need to provide receivers with reasonable feedback on the cost that their network usage incurs. Therefore, cost sharing methods will play an important role in multicast-type services in the future [2] - [4]. There are some types of multicast-type services already in existence. For example, conferencing, contents delivery or video streaming services. In the case of conferencing and contents delivery, we know the number of receivers at the starting point of services. However, in the case of video streaming services, the event of receiver s join and leave occurs continuously between each streaming program. In this paper, we consider the case of multicast-type video streaming services, where the number of receivers changes dynamically. Flat-rate pricing is unrealistic for pricing video streaming services in telecommunication networks as long as network resources are not plentiful, since there would be no incentive to limit wasteful traffic. Therefore, cost-based measured rate pricing is considered in this paper. To simplify the problem, information costs and fixed costs are not considered here; we only consider costs for network usage. If the price is decided dynamically according to the shared cost, the final price cannot be given to any receiver at the start of the service, as the price changes according to the change in the number of receivers joining the service. In this respect, dynamic pricing is not user-friendly. Thus, a time-based pricing where the unit price is fixed is considered in this paper. The other design principles are as follows [5]: receivers do not pay any extra for bandwidth for receiving multicast versus unicast the sender pays for the multicast bandwidth the sender might charge the receivers for the content including the cost above the network providers should receive revenue based on the proportion of their network resources that are used for multicast First, we examine some cost sharing methods that determine charges in multicast-type services and then calculate the probability distribution of the shared cost to analyze how the costs are shared among receivers in the cost sharing methods. II. Cost in Multicast-type Services A. Definition of Cost In this paper, we consider the case where the information provider provides receivers with multicast-

2 type video streaming services by using the network resources that a telecommunication carrier owns. The carrier charges the information provider for its network usage, and this charge is the cost for the information provider to provide a multicast-type video streaming service. Here, information costs and fixed costs are not considered in order to make the problem simpler. B. Cost-based Multicast Pricing The accounting architecture among telecommunication carrier, information provider, and receivers is shown in Fig. 1. The carrier charges the information provider for its network usage. The information provider charges the receivers for contents and network usage for delivering the contents. Hereafter, the charge for contents is not considered because it does not depend on the state of network usage. III. Cost Sharing Methods In multicast-type services, many receivers share a single data flow, as shown in Fig. 2. This section defines three typical methods of cost sharing among receivers for multicast-type service. A. Equal Total Cost Split (ETS) The Equal Total Cost Split (ETS) method is the simplest method for sharing the total cost of all links e- qually among all receivers joining the service. The ETS is a desirable method because of its ability to realize simple accounting and billing. However, the method does not discriminate between receivers far from the sender and those close to the sender. Therefore, the method does not attempt to hold receivers accountable for the costs that their individual membership incurs. Moreover, sometimes a receiver s share of the cost in the multicast-type service becomes higher than that of in unicast service, for instance in the case where some receivers are connected directly to the sender s node and there exists at least one receiver in another node. Therefore, this method does not satisfy individual rationality. B. Equal Link Cost Split (ELS) In the Equal Link Cost Split (ELS) method, the cost of a particular link is incurred because there is at least one downstream receiver. Consequently, while all downstream receivers can be considered equally responsible for the cost, all of the other receivers are not responsible at all. The ELS is an approach where the cost of each link is split equally among only the downstream receivers. The method attempts to hold receivers accountable for the costs their individual membership incurs. Moreover, in this method, multicast costs are always lower than unicast costs. Therefore, this method satisfies individual rationality. However, the cost shared charge for usage requests requests Information Carrier Receivers Provider contents contents charge for information and network usage Fig. 1. Costs and profits for information provider. Sender Receiver Fig. 2. Many receivers share a single data flow. by one receiver can be different from that of other receivers and, as a result, accounting and billing will be complex. C. Equal Group Cost Split (EGS) In the Equal Group Cost Split (EGS) method, all n- odes are divided into m groups, and the total cost the group incurs is equally split among receivers in the group. The EGS substantially has the properties between ETS and ELS, including both their good points and weak points. D. Analysis ETS does not discriminate between receivers far from the sender and receivers close to the sender. Therefore, the method does not attempt to hold receivers accountable for the costs their individual membership incurs. Sometimes a receiver s share of the cost in the multicast-type service becomes higher than that of a unicast cost, for instance in the case where some receivers are connected directly to the sender s node when there exists at least one receiver in another downstream node. However, the tariff is shown to each receiver at the starting point of the service and each receiver is able to decide whether to use a service or not. Moreover, there are multiple senders nodes distributed in the networks and, generally speaking, every receiver can receive the benefit of multicast service if it is seen on the average. However, the realization of the system is its most important property. From this point of new, ETS is a desirable method because of its ability to realize simple accounting and billing. The ELS always satisfies lower multicast prices than unicast prices with reasonable feed back on the cost

3 that their network usage incurs, and satisfies individual rationality. When ELS is used, however, a price is set up which reflects the distance of the sender from the node. As a result, accounting and billing will be complex and the price for similar contents will occasionally change. In this respect, ELS is unrealistic to realize. Moreover, in ELS, the price changes greatly between each program according to the position of each sender and each receiver. This is not user-friendly for receivers to join. In the case of EGS, a price changes according to the distance of the sender from the group. The EGS method is realistic in the case of a multiple carriers model, such as in the case of multiple ISPs model (Fig. 3) because we can regard each ISP as an individual group. Note that there are many studies in which a network usage prices are set up according to the number of passed ISPs. Because each receiver belongs to each ISP, it is unrealistic for each receiver to split the cost with receivers who belong to other ISPs [6]. To summarize, ETS is realistic in the case of a single carrier model, and EGS is realistic in the case of a multiple carriers model. IV. Simulation Model A. Traffic Model In this model, the program is broadcast continuously, and there is neither beginning nor end of the program. To make the calculation simple, it is assumed that an arriving call follows a Poisson distribution and that the holding time follows an exponential distribution. B. Approximative Traffic Model In multicast technology, a point-to-multipoint (multicast) connection is used to copy packets only at branching nodes, which ensures network efficiency. Each node has the ability to copy packets and to forward the copies to receivers or other nodes. The state of each node depends on whether at least one downstream receiver exists. The state of each node is on-state if at least one downstream receiver exists. The state of each node is off-state if one downstream ISP 1 NSP X ISP 2 ISP 4 or IX Sender ISP 3 Receiver Fig. 3. Multiple ISPs model. receiver does not exist. Therefore, it is not the behaviour of each receiver, but the on-off pattern of each node that is used in the simulation to reduce the calculation time [7]. Let N and λ 0 denote the number of receivers connected to each node and the arrival rate of each receiver, respectively. Then, the holding time of the off-state follows the exponential distribution of mean 1 value Nλ 0. The approximative model is used for calculating the holding time of the on-state. The average holding time of on-state E is given as E = (a +1)N 1, (1) λ 0 N where a = λ0 µ 0, and µ 0 denotes the service rate of each receiver. Accordingly, the service rate of each node µ n is given as µ n = 1 E. The network has many nodes and these nodes contain many receivers. The simulation time would be astronomical if the behaviours of all receivers were individually monitored. Therefore, we use an approximative model to calculate the cost (per second per user), as shown below. Let C and C n denote the cost per second per receiver and the cost per second per node in the case of ETS. Let C q and C n,q denote the cost per second per receiver connected to node q and the node q s cost per second share in the case of ELS and EGS. C = C n an, (2) C q = C n,q an. (3) Each node accommodates 500 users (N = 500) in the simulation. C. Network Model A network was modeled as a random graph possessing some of the characteristics of an actual network [8]. The vertices representing nodes were randomly distributed on a rectangular coordinate grid, and each vertex had integer coordinates. For a pair of vertices, say u and v (0 u, v < 1), an edge was added according to the following probability: P e (u, v) = kē v) β exp{ d(u, }, (4) n Lα where n is the number of vertices in the graph, ē is the mean number of degrees of a vertex, k is a scale factor related to the mean distance between two vertices, d(u, v) is Euclidean distance between vertices u and v, L is the maximum distance between any two vertices in the graph, and α and β are the parameters (real numbers between 0 and 1). In this paper, we use two types of random networks to confirm whether the simulation results depend on the network model. These parameters are shown in Table 1. A source node that contains a sender is selected randomly.

4 Table 1. Parameters used for generating random networks. Parameter Value n 50 α 0.25 β 0.20 ē 0 k 25 D. Routing Policy We use simple algorithm, because the routing algorithm is not the main topic in this study. The main topic is the cost-based pricing strategy. We use a shortest path tree (SPT) by employing Dijkstra s algorithm. First, find the shortest path from a source node (in this paper, the node that contains the sender) to all the nodes in the multicast group. Then the shortestpath tree is spanned from a source node to every node in the multicast group. The multicast tree is obtained by deleting nonessential edges. V. Simulation Results A. Cumulative Probability and Costs The relationships between the cumulative probability and the cost (per second per user) are shown in Figs. 4, 5 and 6. Figure 4 shows the result in the case of the Equal Total Cost Split Method. In the case of Equal Link Cost Split Method, costs are calculated and plotted about each node; it is not shown because there would be too many lines. Figure 5 shows the results in the case of the Equal Group Cost Split Method (m = 7), where all nodes are divided according to the hop-count from the sender s node. Figure 6 shows the result in the case of the Equal Group Cost Split Method (m = 2), where all nodes are divided into two groups. Group 1 is the group of nodes in the area where hop-count from the sender s node is less than [ Hmax 2 ]. Here, H max is the maximum hop-count from the sender s node in the multicast tree. Group 2 is the group of nodes in the area where hop-count from the source node is above [ Hmax 2 ]. For example, in Fig. 4, when the information provider sets the probability of having a deficit at 10%, the cost shared by each receiver should be determined as Of course, these values change if the network model or the traffic condition changes. Therefore, we have to predict a traffic load and calculate the cost for each network instead of changing the price dynamically according to the number of receivers between each streaming program. B. Cost Structure Figure 7 shows the relationship between the total costs and the traffic load a in the case of multicast and u- nicast. By using multicasts, total costs can be reduced Fig. 4. Equal total cost split (a =0.002). hop: 0 hop: 1 hop: 2 hop: 3 hop: 4 hop: 5 hop: 6 Fig. 5. Equal group cost split (m =7,a =0.002). Group 1 (Close to sender's node) Group 2 (Far from sender's node) Fig. 6. Equal group cost split (m =2,a =0.002). drastically. This benefit will be allocated to receivers, carriers, and senders based on a certain economic law. The relationship between the total costs and the number of nodes is shown in Fig. 8. The total costs increase in proportion to the increase in the number of nodes. The relationship between the total costs and the traffic load a is shown in Fig. 9. As traffic load a increases, the total costs increases. Though the total costs increase with the rise in traffic load a, the value of the total costs is soon saturated. Therefore, the total cost is a constant value when the traffic load is high. Conversely, when the traffic load is low, we have to predict the traffic load and the total cost precisely.

5 Total cost x Unicast Multicast Traffic load 'a' Fig. 7. Total costs, multicasts vs. unicasts. Total cost x a=0.001 a= Number of nodes Fig. 8. Total costs and the number of nodes. Total cost x Traffic load 'a' 0.01 Fig. 9. Total costs and traffic load a (number of nodes: 50). VI. Conclusion In this paper, three cost sharing methods for multicast-type services were defined. By using these methods, the charge for a multicast-type video streaming service can be determined in such a way as to reflect the cost that each receiver s network usage incurs. By analyzing the methods, we found that the Equal Total Cost Split method is realistic in the case of a single carrier model and that the Equal Group Cost Split method is realistic in the case of a multiple carriers model. The shared costs in the above methods were calculated in the simulations. We verified that the multicast-type services were rational. We found that the total costs increase in proportion to the increase in the number of nodes. We also found that the values of multicast costs depended on the traffic load. Though the total costs increase with the rise in traffic load, the value of the total costs is soon saturated. In other words, the total cost is a constant value when the traffic load is high. Conversely, when the traffic load is low, we have to predict the traffic load precisely and calculate the cost for each network instead of changing the price dynamically according to the number of receivers at each point in time. Though we mainly focused on algorithmic aspects, we are aware that a number of issues still remain unresolved, including users behaviour in response to tariffs, and a traffic forecasting method. Acknowledgements This study was partly supported by a Waseda University Grant for Special Research Projects (Individual Research: 2000A-926). Part of this study is from the results of the project Research on Pricing Controlled Highly Efficient Network at Advanced Research Institute for Science and Engineering, Waseda University. References [1] T. Henderson and S. Bhatti, Protocol-independent multicast pricing, The 10th International Workshop on Network and Operating System Support for Digital Audio and Video, June [2] K. Ravidran and T.-J. Gong, Cost Analysis of Multicast Transport Architectures in Multiservice Networks, IEEE/ACM Transactions on Networking, Vol. 6, No. 1, pp , February [3] S. Herzog, S. Shenker and D. Estrin, Sharing the Cost of Multicast Trees: An Axiomatic Analysis. IEEE/ACM Transactions on Networking, Vol. 5, No. 6, pp , December [4] J. Chuang and M. Sirbu, Pricing Multicast Communication: A Cost-Based Approach, Proceedings of the Internet Society INET 98 Conference, Geneva, Switzerland, July (retrieved Dec. 9, 2000 from sirbu/pubs/98354/chuang.html) [5] S. Kasera, M. Hofmann and R. Miller, A Profitable Multicast Business Model, New York Metro Area Networking Workshop, Hawthorne, New York, USA, March 12, [6] N. Ogino and M. Suzuki, Proposal of a Price-Based Inter- AS Policy Routing to Improve ASes Profits, IEICE Trans. Commun., Vol. E85-B, No. 1, pp , January [7] Y. Tanaka and R. Nishino, Modeling of Broadcast Viewing Traffic (in Japanese), The Transactions of IEICE, vol. J78- B-I, No. 4, pp , April [8] B. Waxman, Routing of Multipoint Connections, IEEE Journal on Selected Areas in Communications, Vol. 6, No. 9, pp , December 1988.