Streaming Variable Bit-Rate Media Streams - Scalable Continuous Media Streaming Systems - page 121

of the system's performance. For example, McManus and Ross [2] suggested dividing the
entire video stream into fixed size intervals and then transmitting each interval with a constant
bit-rate to enable control of the rate adjustment frequencies. Feng et al . investigated smoothing
algorithms to minimize the number of rate increases [3], and to minimize the number of rate
changes [4]. In another two studies by Feng [5, 6], the author observed that in some cases, an
algorithm targeted at minimizing a certain parameter might make too aggressive prefetches or
allow too large buffer residency times. The author then proposed a rate-constrained smoothing
algorithm [5] and a time-constrained smoothing algorithm [6] to solve these problems. Chang
et al . suggested that transmitting at a constant rate yields lower overhead and complexity.
Therefore, they proposed a smoothing algorithm that switches a single constant transmission
rate on and off to adapt to the video bit-rates [7]. Salehi et al . investigated the optimal smoothing
algorithm [8] that produces smoothing schedules with minimum peak rates and rate variations.
Interested readers are referred to the excellent study by Feng and Rexford [10] for a detailed
survey and comparison of various smoothing algorithms.
Besides smoothing algorithms based on finding a path inside the feasible region, there are
also other related studies in this area. For example, Zhang [9] proposed smoothing using
buffers located in multiple intermediate nodes in the network. Zhao and Tripathi [11] proposed
an algorithm to multiplex smoothed VBR streams to further reduce bit-rate variations. Liu
et al. [12] observed that scene changes in a video usually correlate with bit-rate variations, and
thus proposed an algorithm to detect scene changes to allocate a constant bit-rate for each scene.
For real-time videos, Rexford et al . [13] proposed an online, lossless smoothing algorithm that
uses a sliding window. Liew and Tse [14] proposed using client buffer occupancy to control
encoding parameters for smoother encoder output. In another study, Duffield et al . [15] used
network status feedback to control the encoding parameters.
7.2 Streaming in Mixed-Traffic Networks
After smoothing is performed, the transmission schedule of a VBR video will be reduced
to a series of constant-rate segments (see Figure 7.1). The media server can then reserve
bandwidth for these segments before transmitting them over the network to the client. As
long as the bandwidth reservations are successful, timely delivery of the video data to the
client can be guaranteed. However, two factors in practice often affect the effectiveness of this
approach.
First, although the bit-rate of each smoothed segment is constant, the system still needs
to successfully complete the bandwidth reservation process before the next segment can be
transmitted. The adjustments needed may contain both downward adjustments (switching from
a higher bit-rate to a lower bit-rate), and upward adjustments (switching from a lower bit-rate to
a higher bit-rate). The former case is straightforward as less network resource will be required,
but the latter case is more complicated. In particular, if the network concurrently carries traffic
from other applications (e.g., Web, FTP, other video streams, etc.), it is conceivable that the
upward adjustments could fail when the additional bandwidth is not available at that moment .
Clearly this will result in either disruption of the video stream or severe quality degradation
such as playback jitter. As the instantaneous bandwidth consumption in a network with mixed
traffics is inherently unpredictable, this problem is unavoidable unless one dedicates a portion
of the network resources to a video stream. However, this clearly will result in significant
over-engineering and thus defeat the whole purpose of smoothing in the first place.