Quality of Service for Internet Multimedia: a General Mapping Framework
3.1 Introduction
Ongoing research efforts regarding service differentiation can be classified into two approaches: absolute differentiated service and relative differentiated service, as mentioned in Section 2.5.
Absolute differentiated service assures an admitted user of the promised performance level of the DiffServ classes, independent of the traffic status of the network. Therefore, such an approach is useful for applications that require strict QoS guarantees. On the other hand, realization of absolute differentiated service requires stringent admission control.
On the application side, much intelligence can be added to application-layer protocols nowadays. In the case of a multimedia application, the transmission rate can adapt to network congestion through, for example, a choice of different compression rates. Regardless of absolute or relative differentiation, the DiffServ framework can best be utilized by implementing some intelligence at the boundary nodes of the DiffServ domain. As CM applications become more network-adaptive, it appears that many will perform well with relative differentiated service, even though the network performance guaranteed by relative DiffServ is not as solid as that guaranteed by absolute DiffServ. Except for several conversational CM applications (e.g., Internet telephony, including video conferencing), the majority of networked CM applications are tolerant of occasional delay/loss violations. Thus, they do not require tight delay/loss bounds, which can be better provided by DiffServ premium service (PS).1 For streaming video applications, encoding/decoding is more resilient to the loss rate and delay fluctuations, and thus the capability of relative service differentiation seems adequate for the streaming video applications on which we will focus.
With an appropriate pricing rule, the method of exploiting relative differentiation for CM applications seems to be an important issue for successful cooperation between the differentiated service and the CM applications. Thus, in this chapter, we propose a subscription-based pricing model for differentiated service quality among DS levels2 specified in the SLA, under which there is a futuristic framework for QoS mapping between practically categorized packet video and a relative DiffServ network employing a unified priority index and an adaptive packet-forwarding mechanism. In this framework, the video application at the source grades the chunks of its content by certain indices (i.e., categories for packets) according to their importance in end-to-end QoS (e.g., in terms of loss probability and delay). Since these indices reflect the desired service preference of a packet compared with others in fine granularity, we denote them with an RPI, which is further divided into a relative loss priority index (RLI) and a relative delay priority index (RDI).
Next, QoS control takes place through the assigning of an appropriate DS level to each packet, a process that we call QoS mapping. Using the RPI association for each packet, an efficient (i.e., content-aware) mapping can be coordinated either at the end-application or at the boundary node. Note that the efficiency of QoS mapping for relative differentiation is dependent on the persistence of the contracted (advertised) quality differentiation over different time scales in the presence of traffic fluctuation. That is, the packet-forwarding mechanisms (e.g., queue management and packet scheduling) of DiffServ needs to provide the target performance differentiation persistently over time. (In Chapters 5 and 6, we will discuss how to seek persistent service differentiation and how such persistence improves end-to-end quality through QoS mapping.) With a relative DiffServ network providing consistent service differentiation persistently over time, CM applications, including streaming video, can be built more reliably and for less cost.
This chapter presents a relative service differentiation framework connecting CM applications, especially streaming video applications, through the proposed RPI. The chapter addresses the following issues: (1) a relative priority-based, per-packet video categorization in terms of delay and loss; and (2) an optimal (or effective) QoS mapping between application categories and DS levels under the pricing cost constraint of the relative service differentiation network. Actually, this framework belongs to joint source/channel coding, and more specifically to the UEP technique. Commonly, UEP enables prioritized protection for source layers (e.g., layered streams of video). It can be realized at the transport end with different levels of FEC and/or ARQ for each layer [91], [92], [93], [94]. However, to the best of our knowledge, no UEP approach has addressed the issue of using packet-level, fine-grained prioritization of the proposed RPI instead of layered protection. For a DiffServ network especially, only layered prioritization in an absolute differentiation sense has recently been proposed in [95], utilizing the video object layer of MPEG-4 and a different packet-discarding mechanism.
The rest of this chapter is organized as follows. The proposed QoS mapping framework is described in Section 3.2. Video categorization with RPI according to several criteria is examined in Section 3.3 for the case of ITU-T H.263+ video [96]. By investigating the error-resilient version of the H.263+ stream, the RPI is assigned so that different video packets can be tied to the relative loss rate/delay differentiation of DiffServ networks. Then, optimal QoS mapping guidance is presented for a certain packet loss rate and given cost curve according to the DS level. Performance assessments using the random loss pattern from a two-state Markov model for each DS level and H.263+ video are given in Section 3.5, where the implications of experimental results are also discussed. Finally, concluding remarks and anticipated future research efforts are given in Section 3.6.