Taxonomy and Architectural Alternatives - Scalable Continuous Media Streaming Systems - page 159

9

Taxonomy and Architectural
Alternatives

In Part I we covered the basic principles and concepts in media streaming based on the
client-server model. Two issues in particular recur frequently in the discussions - capacity
and reliability. Capacity dimensioning is crucial in a media streaming system as predictable
performance is often needed to sustain continuous media playback at the clients. When
the maximum server capacity is reached, the server will need to deny new user requests
so that the performances of existing streams are not adversely affected. As discussed in
Chapter 3, we can expand the server's streaming capacity by employing striped disk array
and multi-disk scheduling. Nevertheless, we cannot keep adding more disks to an existing
media server to expand the streaming capacity because the server will eventually run into
other bottlenecks in the server host, such as the I/O bus capacity or CPU processing limit.
Thus, the capacity of a single media server is still ultimately limited.
On the other hand, reliability is also an important issue in practice, especially when
providing paid streaming services to a large user population. Within the media server
we can improve reliability by using technologies such as error-correcting memory chips,
redundant power supplies, and RAID storage. However, failure at the server level, such as
those caused by software bugs, hardware failure, and so on, will not be recoverable. Current
solutions such as stand-by systems can be used to restore the system but in addition to the
high costs of fully-replicated hardware, the on-going streams will also likely need to be
restarted in the process of switching to the stand-by systems.
The previous discussions illustrate the limitations of the single-server architecture in im-
plementing media streaming services. In Part II we depart from this architecture to investi-
gate an alternative: parallel server architectures, that offer promising solutions to the above-
mentioned scalability and reliability limitations. In this chapter we give an overview of the
issues in designing streaming systems around parallel server architectures, and present a tax-
onomy to classify the many possible architectural alternatives. Subsequent chapters will in-
vestigate in more detail several architectures which provide different engineering tradeoffs.