QoS enabled Layered Multicast Transmission of Multimedia Data in a Wireless Environment
B.S. Srinivas Hemant Chaskar Dirk Trossen (bindignavile.srinivas@nokia.com) (hemant.chaskar@nokia.com) (dirk.trossen@nokia.com) Nokia Inc. 5, Wayside Road, Burlington, MA 01803 USA
Abstract: The ability to transmit multimedia data to mobile hosts is becoming increasingly important in the wireless domain. In the context of transmitting the information to a multiplicity of hosts, efficiency and cost factors dictate the use of a multicasting approach rather than multiple unicast transmissions. Furthermore, the existence of user terminals with varying display capabilities as well as the varying bandwidths of different wireless segments necessitates the use of a layered transmission approach. In this work, a single multicast group address is used for all the layers of encoded multimedia data. Each encoded layer is placed in a distinct IP packet stream. The IP packets carry priority information of the corresponding stream in the DS (Differentiated Services) field, which corresponds to the layer represented by the packet payload. Using the priority information, in the presence of congestion, the multicast enabled router makes a forwarding decision for each packet. A multicast extension to mobile IP (MIP) is employed to enable multicasting in a mobile environment.
1. INTRODUCTION: The current manifestation of the Internet is as a heterogeneous set of links of varying capacities (bandwidths) and a population of end devices (terminals) with a multiplicity of reception capabilities. Among the numerous feasible applications being investigated for the modern Internet, the multipoint transmission of multimedia data is an important one. A trivial solution for the task of multipoint distribution over a heterogeneous Internet is to distribute a uniform representation of the signal to the set of receivers using IP Multicast [1]. This approach, however, is suboptimal, with low capacity links becoming congested while high capacity ones are underutilized. Alternatively, in a unicast setting, a rate adaptive source transmission approach may be implemented which reacts to link capacity heterogeneity as well as network congestion [2, 3]. However, in a multicast setting, varying the rate of a single stream, to meet the conflicting requirements of the whole population of receivers, is an impossible task. Instead, an approach suggesting the combination of a layered compression algorithm with a layered transmission scheme [4, 5] was proposed. The approach suggests encoding the signal into multiple layers, which can be additively combined to achieve progressive refinement, coupled with the ability to selectively drop layers at various points in the network based on static as well as dynamic network conditions (i.e. congestion). The rest of the paper is structured as follows. Section 2 describes some related work while Section 3 delves into details of the proposed technique. Finally, Section 4 discusses the conclusions of this work. 2. RELATED WORK: Hierarchical or layered coding [4], a family of signal representation techniques, refers to an approach in which the source information is partitioned into sub-streams or layers, each representing a portion of the signal. The greater the number of layers received at the end stations, the better is the quality of the reconstructed signal. Using explicit control signals from the receivers, a rate-control based approach may be used to deal with the issue of congestion [3]. However, source-based rate control is a bad fit with multicasting to heterogeneous receivers. Seminal work in the area of IP multicasting was carried out by McCanne [6]. The work proposed striping the different layers of the layered signal across multiple multicast groups followed by the adapting of receivers to congestion scenarios by adding and dropping layers (i.e. joining and leaving multicast groups). In other words, each layer of the source material is assigned to a separate multicast group, following which, based on their capability as well as current network conditions, receivers join and/or leave the different groups. Receivers implicitly define the multicast distribution trees by expressing their interest in receiving flows. In contrast to the approach proposed in [6] where each encoded layer is transmitted on a different multicast group, the current work proposes using a single multicast tree for all the layers. Each receiver makes an independent decision to join or leave a particular multicast tree based on its rendition capability as well as local congestion status. In the proposed work, the router examines the DiffServ (DS) field in the IP header to learn the layer, which a particular packet belongs to and then makes a decision about whether the packet should be forwarded to the downstream node or dropped instead. The desire to receive a particular layer as well as less important ones is based on the receiver’s capability as well as local congestion. The multicast distribution tree, in the proposed approach, is constructed based on explicit ICMP (actually IGMP messages which has been incorporated into ICMPv6) messages being transmitted from the leaves (receivers) to the router. The proposed approach is similar to that presented in [7] with the difference being that the proposed work recommends using the class field in the IPv6 header for layer identification, while the latter suggests using an extension to the RTP header. 3. QOS ENABLED LAYERED MULTICAST: (a) Congestion Location: The multicast scenario envisages the occurrence of congestion at two alternate locations. Namely, congestion may occur either at the router (point of duplication of the packet) in the multicast distribution tree, or it may occur at the individual leaves (receivers) of the tree. Congestion is typically detected by the loss of packets in the data stream (ECN queuing uses queue levels to detect the onset of congestion rather than packet loss). In either of the two scenarios described, receiver-driven layered multicast (RLM) [6] may be used to alleviate congestion. Essentially, RLM guides the addition or deletion of encoded layers based on the detection of congestion. However, a pertinent difference between the two cases is that while RLM, when utilized at the receivers, has a low latency in terms of the impact on the congestion experienced by the receivers, RLM, when utilized to reduce congestion at the router has a larger latency and hence a lesser impact on congestion mitigation. Given the penalty of a higher latency near the root (of a multicast distribution tree) as described above, the goal of this work is to alleviate the latency cost without losing the benefits of RLM. In order to do so, the ability to differentiate between the different source layers is needed. This can be achieved, for instance, by using IP addresses. However, since IP addresses are not ordered according to the priority of the media layer, some other parameter (such as port number), which is ordered, is needed to differentiate between the layers. (b) Use of DiffServ Field: Given the need for a technique to differentiate between the different layers of the encoded bit-stream, we propose using the "class" field of the IP header to carry the (layer) identification information. With the rapid growth of the Internet and the limited supply of IP addresses available from the older IP protocol, IPv4, the need for a new protocol with a much larger supply of addresses was felt. The new protocol, IPv6, with an IP address field of 128 bits as against the older IPv4 address field length of 32 bits fits this need. In addition to the increase in the number of IP addresses, the structure of the IP header has also been streamlined with a far fewer number of fields in IPv6. However, the IP header has expanded from 20 bytes (in IPv4) to 40 bytes (in IPv6). A pictorial representation of the IPv6 header structure is shown in Figure 1.
Figure 1: IPv6 Header The "Class" field in the IPv6 header serves as the "priority" field, which could be used as the DS field. In particular, the different layers of the hierarchically encoded source are each carried on separate IP packets, which can each, be assigned different (layer based) values in the DS field. The value in the DS field is examined by the router (at the root of the multicast tree), which then makes a decision to forward or discard an IP packet to a particular leaf. (c) Multicast extension to Mobile IP (MIP): Mobile IP (MIP) proposes two approaches for supporting multicast to mobile hosts [8, 9] namely remote-subscription (foreign agent based) and bi-directional tunnelling (home agent based). The former approach is a very simple scheme without any requirement of encapsulation, instead requiring a mobile host (MH) to register with the foreign agent (FA) and issuing a binding update to its home agent (HA) updating its co-located address. This approach has the benefit of more efficient shortest path routing. However, its primary demerit is the overhead costs associated with frequent distribution tree updates resulting from MH mobility. Conversely, in the latter approach, each multicast datagram is delivered to a MH by way of its HA, which results in routing inefficiency due to triangle routing [9]. The multicast delivery tree does not have to be frequently updated (due to host mobility) resulting in a saving of the associated overhead. 4. CONCLUSIONS:
We have investigated an alternative approach to multicast multimedia data using a single multicast distribution tree. This technique has a lower ICMP packet overhead since it sets up a single multicast distribution tree. Conversely, the traditional RLM technique [6], which sets up multiple multicast trees, one for each layer, has a higher tree construction overhead. A hierarchical encoding with layer based IP packet payload as well as DS field marking has been used. A brief discussion on the multicast extensions to MIP has also been presented. 5. REFERENCES:
|