Introduction - Scalable Continuous Media Streaming Systems - page 21

algorithm used will affect the resultant media bit-rate. Take video compression as an example.
There are many video compression algorithms available, such as MPEG2, MPEG4, and H.264.
In terms of compression efficiency, i.e., achievable compression ratio at a given video quality
level, it is generally agreed that MPEG2 H.264. However, the more efficient video

codec such as H.264 also demands substantially more computations in both the encoding and
decoding processes. Thus, choosing a more efficient compression algorithm such as H.264
will lower the network bandwidth required but will increase the computation complexity at
both the encoder and decoder.
Note that for stored media the encoder complexity is less of an issue as the encoding process
can be performed offline. However, if the media stream is encoded from a live source in real-
time, then the encoding complexity will become a significant constraint. Decoding, on the
other hand, is usually less complex than the encoder and thus presents less of an issue in the
choice of codec in a media streaming system.

< MPEG4 <

1.7 Performance Guarantee
In the previous discussions we illustrated five common dimensions for engineering trade-off.
Regardless of the approach taken, the goal is to deliver the media data to the client in time for
playback, i.e., to provide performance guarantee. Common to all these different approaches is
the need to consider the worst-case scenario - in this case the peak bit-rate among all media
segments. This is typical in designs that provide deterministic performance guarantee, i.e., the
performance is met under all valid scenarios.
It is easy to see that deterministic performance guarantee often result in poor resource
utilization, especially if the worst-case scenario rarely occurs. Alternatively, if we relax the
requirement to meet performance under all valid scenarios, and instead guarantee that perfor-
mance ismet most of the time, thenwe can often reduce the resource requirements substantially.
This approach is often known as probabilistic performance guarantee or statistical performance
guarantee. The trade-off then is between resource utilization and the probability of failing the
performance guarantee.
Finally, we should also mention a third type of performance guarantee (or lack thereof)
- best effort. By best effort it implies that the system will attempt to meet the performance
requirements using the available resources but there is no guarantee at all. Note that best effort
does not necessary mean poor service. Rather, it simply implies that the probability that any
given performance requirements is met is not known or controllable.

1.8 Admission Control
An issue closely related to providing performance guarantee is the need for admission control.
Specifically, when we consider the different engineering tradeoffs in Section 1.6 we have
always assumed that a given network bandwidth needs to be allocated before media streaming
can start. In other words, if the network utilization is so high that the required bandwidth is
not available, the media server will then reject the request for a new media streaming session
(Figure 1.20). This process is known as admission control and it is one of the key elements in
providing performance guarantee.