Introduction - Scalable Continuous Media Streaming Systems - page 10

Compared to the download model, there are now two additional requirements for stream-
ing to work. First, the media object must be decomposable into smaller fragments that are
independently or progressively (i.e., making use of the current and already received frag-
ments) decodable and presentable. Most continuous media such as audio and video pos-
sess this property. Second, to ensure the timing integrity in presenting the media object,
we will need to ensure that each and every media fragment can be delivered to the client
before their scheduled playback time. This is also known as the continuity requirement and
is one of the key performance metrics used in designing and evaluating continuous media
systems.
With the rapid advances in networking technology one may wonder if in the future our
networks will be equipped with so much bandwidth that the transmission time will become
insignificant, even using the download model. This is indeed a valid question but in addition
to allowing playback to begin earlier, the streaming model also offers another significant
advantage - pipelining of multiple concurrent streams.
Specifically, media servers often need to serve many clients concurrently. When multiple
clients request service at around the same time, there are two options for the media server if
the download model is used - it can either serve the clients one after the other in a sequential
manner, or it can serve them simultaneously. In the former, all but the head-of-line client will
need to be queued and thus experience additional queueing delay when waiting for service. If
the download model is employed, then even in a very high bandwidth network the waiting time
will still be significant. In the latter case, serving multiple clients concurrently will reduce the
network bandwidth available to each client and thus increases the download time proportionally.
By contrast, the streaming model does not suffer from this problem as playback can begin
once an individually decodable fragment of media data is received (Figure 1.8). Moreover, as
the media data are often transmitted at the media playback data-rate, the start-up delay is in fact
independent of the number of clients requesting service simultaneously, as long as the media
server and network capacity are not exceeded. This unique multistream pipelining property
can significantly reduce start-up delay especially during high system utilizations.

Client #1

Server

. . . . . . . . . . . . . . . . . . . . . . . .

Client #2

Figure 1.8 Multistream pipelining in the streaming model