Layered Multicast Of Scalable Media : Introduction

With the fast deployment of Internet infrastructure, wired or wireless, the IP network is getting more and more heterogeneous. The heterogeneity of receivers under an IP multicast session significantly complicates the problem of effective data transmission. A major problem in IP multicast is the sending rate chosen by the sender. If the transmitted multimedia data rate is too high, this may cause packet loss or even a congestion collapse whereas a low transmission data rate will leave some receivers underutilized.

Heterogeneity of the receivers under an IP multicast session.

This problem has been studied for many years and is still an active research area in IP multicast. To solve this issue, the transmission source should have a scalable rate, i.e., multirate, which allows transmission in a layered fashion. By using multirate, slow receivers can receive data at a slow rate while fast receivers can receive data at a fast rate. In general, multirate congestion control can perform well for a large multicast group with a large number of diverse receivers. This brings us to the scheme of layered multicast.

Basically, layered multicast is based on a layered transmission scheme. In a layered transmission scheme, data is distributed across a number of layers which can be incrementally combined, thus providing progressive refinement. The scalable video coding (SVC) can easily provide such layered refinement. Thus, the idea of layered multicast is to encode the source data into a number of layers. Each layer is disseminated as a separate multicast group, and receivers decide to join or leave a group on the basis of the network condition. It is assumed that the data that is going to be transmitted can be distributed into l multicast groups with bandwidths Li, i=0, . . . ,l-1 . Receivers can adjust the transmission rates by using the cumulative layered transmission scheme. Now the adaptation to heterogeneous requirements becomes possible because it can be done independently in each receiver. On the basis of the network condition, a particular receiver can subscribe a bandwidth Bi by joining the L0, L1, . . . , Li layers:

Bandwidth formula on layered multicast group.

The more layers the receiver joins, the better quality it gets. As a consequence of this approach, different receivers within a session can receive data at different rates. Also, the sender does not need to take part in congestion control.

Layered multicast group.

To avoid congestion, end systems are expected to be cooperative by reacting to congestion and adapting their transmission rates properly and promptly. The majority traffic in the Internet is best-effort traffic. The transport control protocol (TCP) traffic uses an additive-increase multiplicative-decrease (AIMD) mechanism, in which the sending rate is controlled by a congestion window. The congestion window is halved for every window of data containing a packet drop and increased by roughly one packet per window of data otherwise. Similarly, IP multicast for UDP traffic needs a congestion control algorithm. However, IP multicast cannot simply adopt the TCP congestion control algorithm because acknowledgements can cause an “implosion problem” in IP multicast. Owing to the use of different congestion control algorithms in TCP and multicast, the network bandwidth may not be shared fairly between the competing TCP and multicast flows. Lack of an effective and “TCP friendly” congestion control is the main barrier for the wide-ranging deployment of multicast applications.

“Scalability” refers to the behavior of the protocol in relation to the number of receivers and network paths, their heterogeneity, and their ability to accommodate dynamically variable sets of receivers. The IP multicasting model provided by RFC-1112 is largely scalable, as a sender can send data to a nearly unlimited number of receivers. Therefore, layered multicast congestion control mechanisms should be designed carefully to avoid scalability degradation.