Introduction - Scalable Continuous Media Streaming Systems - page 14

determining the service's cost-effectiveness. Many early continuous media streaming services,
video-on-demand in particular, have failed to reach wide acceptance partly because the cost
of provisioning the service is too high. Along with the recent advances in processor, storage,
and network technologies, the cost in storing and delivering vast amount of media data has
dropped substantially. Nevertheless, the cost in serving high-quality media contents such as the
emerging high-definition video to a large number of users (e.g., in a city) are still very substantial
and thus the quest for ever more efficient media streaming system designs continues to be an
important research topic.
1.5.5 Scalability
Another challenge related to efficiency is scalability. Specifically, scalability refers to the limit
at which one can increase the service capacity of a system, and the rate of increase in the system
cost when the system capacity is scaled up. Consider a simple system with one media server
as shown in Figure 1.12a. When more and more clients join the system, the media server will
eventually become overloaded (Figure 1.12b), thus leading to unacceptable waiting time or
service quality. One possible solution is to add a new media server to the system, and replicate
all media contents to the new media server as shown in Figure 1.12c. This doubles the system
capacity at the expense of doubled system cost.
More generally, Figure 1.13 illustrates three types of cost/capacity relations when we scale
up the capacity of a system. In Case #1, the cost per unit capacity increases when one increases
the system capacity. For example, if we increase the capacity of a media server by replacing
it with a higher-capacity server, then it is quite common that the cost per unit capacity will
increase for servers of higher and higher capacity. This is due to the lack of economy of scale
in producing the very high capacity servers compared to the mass-produced commodity server
platforms.
In Case #2 the cost per unit capacity is constant. Our previous example of replicated media
servers falls within this type of scalability. Finally, in Case #3 the cost per unit capacity
decreases with increases in the system scale. This is obviously highly desirable as it implies
that a service operator can benefit from economy of scale in provisioning media streaming
services to a large user population. The multicast streaming architectures to be covered in Part
III of this topic will cover many streaming architectures that achieve precisely this type of
scalability.
1.5.6 Reliability
In addition to scalability, service reliability is another important challenge in provisioning
large-scale media streaming services. Starting from the storage subsystem such as a disk
array, the failure of a disk will disrupt the operation of the media server unless fault-tolerant
mechanisms (e.g., RAID [2]) are employed. In addition to disk failures, the media server itself
is also susceptible to many potential failures, including memory failure (some of which can be
corrected using error-correcting memory chips), network interface failure, processor failure,
power failure, or simply due to hitting a bug in the media server software.