Quantitative performance comparison of different content distribution modes
Introduction
Traditionally, broadcasting has been used for transmitting mass media content. Consider, for example, a TV or a radio transmission, where users can use their receivers to tune to the desired channel. The fragmentation of user preferences and societies leads to the situation where the interests of user groups diverge, and the number of users tuned to a specific channel decreases. In this type of an environment, new technical challenges arise. Namely, the content providers need to provide a wider assortment of mass media transmissions for smaller user groups. In this case, the traditional broadcast transmission mode does not scale well, leading to a waste of transmission capacity. On the other hand, unicast transmission, traditionally used for telephone calls, does not scale well for this application, neither. One solution is using multicast as the method for group communication, where several users may join and listen the same transmission.
The delivery of mass media content to mobile devices in cellular networks is a challenging problem. Delivering content using IP over broadcast radio technologies like Digital Video Broadcasting (DVB-H for mobile devices) is referred to as IP Datacast (IPDC) [1]. IP Datacasting is based on the IP multicasting paradigm [2], with some conceptual additions for unidirectional networks and/or service concepts [3], [4], [5], [6]. Mobile devices receiving IPDC have been demonstrated.
We study the efficiency of different delivery techniques for streaming type mass media contents. As noted, there are three basic transmission modes available: unicast, multicast, and broadcast. In unicast each user requires a dedicated transmission, while in multicast and broadcast a single transmission per content is enough for several users. Thus, the latter two modes are more efficient. On the other hand, in the unicast and multicast modes, the transmissions are activated by users implying that the number and assortment of contents under transmission varies randomly in time, whereas in the broadcast mode the number and assortment of contents remains fixed (for longer periods), since the operator makes the transmission decisions. In addition to these basic modes, we study a combined mode, where the most popular contents are broadcast while the others are transmitted using unicast. We call this mode the broadcast-unicast mode.
The multicast gain within a single multicast group over the whole network was evaluated by Mieghem et al. [7], who studied the average number of joint hops in a shortest path multicast tree. It is also possible to study the gain of group communication in the viewpoint of a single link: how much of the link capacity can be saved for a fixed user population, or how many more users can be supported for a fixed link capacity. Azar et al. [8] studied the gain of using multicast for a web-site, where the most popular pages would be sent using a multicast carousel transmission. They defined the gain as the ratio of required bandwidths in different transmission modes for a fixed traffic load. In [9], we defined the multicast gain as the ratio of the number of users supported in different transmission modes for a fixed grade of service parameter (call blocking probability).
This paper extends the methodology and results presented in [9]. We consider the gain of group communication (multicast and broadcast-unicast) achieved in a single link, such as a cell in a cellular network. This is reasonable due to the limited transmission capacity of the radio interface when compared to the wired core network. The questions we study are:
- •
How to define the gain of group communication?
- •
How do different traffic and system parameters affect the gain?
- •
Which parameter values give rise to a significant gain?
- •
Which transmission mode gives the highest gain under given parameters?
The rest of the paper is organized as follows. Section 2 presents the model and defines an average link occupancy based metric that can be used for quantifying the group communication gain. Section 3 introduces another metric for the group communication gain based on the blocking probabilities, and develops analytic and simulation-based methods to evaluate the gain. Due to some problems on the scalability of the simulation approach, an approximative approach for large systems is developed in Section 4. Numerical results including the comparison of different transmission modes are presented in Section 5. Conclusions are drawn in the final section.
Section snippets
Comparison method based on average link occupancy
This section presents the system and user models, and develops a simple comparison method between the transmission modes based on the average link occupancy in a system with unconstrained capacity.
Comparison method based on blocking probability
In this section we consider the system as a loss system with the capacity constrained to C, see e.g. [13] for an introduction to such models. Calculating the call blocking probability in such a system leads to a more user-centric (but also a more complex) metric for comparison of different transmission modes.
In the unicast mode, given the capacity of a link and the target blocking probability, the maximum number of users can be calculated by inversing the well known Engset formula [14]. A
Comparison method for large systems
This section develops an approximative approach for large systems (with a large number of users and high capacity links). The approximation is based on the large deviations theory (see e.g. [25] for an introduction to this theory). More precisely, we use a Bahadur and Rao [26] type approximation of the time blocking probability, similarly as for the unicast case in [27].
Numerical results
In this section we present the numerical results and, based on these results, a comparison between different transmission modes. The important parameters affecting the results are the shape parameter of the preference distribution, the total number I of different contents, the link capacity C, and the user activity p. All the numerical results are based on the user activity level . Other parameters are varied to get a comprehensive comparison. In particular, it is interesting to see the
Conclusions
We have defined multiple figures for quantifying the gain from the use of multicast or a combination of broadcast and unicast in a single transmission link, such as a cell in a cellular network, and provided the methods for calculating or estimating the figures. For the broadcast-unicast mode, the operator should be aware of the users’ preference distribution, while in the multicast mode, it is not required. The gain figures are consistent with each other, and suggest that, in certain
Acknowledgements
We thank Jarmo Mölsä for his comments that helped us to improve and clarify our presentation. We also thank the anonymous referees for their constructive comments.
References (28)
- et al.
An exact end-to-end blocking probability algorithm for multicast networks
Perform. Eval.
(2003) - et al.
Multicast context delivery for mobiles
- S. Deering, Host extensions for IP multicasting, RFC 1112, August 1989....
- et al.
Providing high-speed Internet access using DVB broadcast channels
- J.-P. Luoma, Internet access over DVB networks, Master’s thesis, Tampere University of Technology (Feb....
- Mobile Internet Technical Architecture, Technologies and Standardization, IT Press,...
- J. Aaltonen, Content distribution using wireless broadcast and multicast communication networks, Ph.D. thesis, Tampere...
- et al.
On the efficiency of multicast
IEEE/ACM Trans. Network.
(2001) - et al.
The multicast bandwidth advantage in serving a web site
- et al.
Multicasting vs. unicasting in mobile communication systems
Web caching and Zipf-like distributions: Evidence and implications
The content and access dynamics of a busy web site: findings and implications
Multiservice Loss Models for Broadband Telecommunication Networks
Cited by (1)
Efficiency of personalized content distribution
2008, Bell Labs Technical Journal