Closed-Loop Algorithms - Scalable Continuous Media Streaming Systems - page 296

2
4
5
8

Incoming
requests

1 1 3 7 3 3 1 3 3 6 7 9 1 9 3 7 3 3 1 3 3 6 7 9 1 9

wait

FIFO Queue

4 -Streams
Media Server

A media stream is multicast to all users requesting the same video.

Figure 17.1 First-come-first-serve batching

way. In multicast media streaming, the batching technique can also be applied to improve the
bandwidth efficiency of multicast, i.e., to increase the sharing of multicast data by more users
[1-4].
Figure 17.1 illustrates the batching principle using a media server with a capacity of four
concurrent multicast channels - a channel represents the retrieval and transmission capacity
to support one multicast media stream (we assume all multicast channels are of the same data
rate). In Figure 17.1 all four multicast channels are busy serving on-going streaming sessions.
As users arrive to find the server busy, theywill queue up in a first-in-first-out (FIFO) queue. The
numbers in the box in Figure 17.1 represents the identity of the media object being requested.
Thus, the four on-going channels are streaming data from media object “2”, “4”, “5”, and “8”
respectively.
Now suppose after some time the media channel streaming media object “8” is completed
and thus becomes available again as shown in Figure 17.2. Then in a unicast-based media
streaming system the server will simply pick the head-of-line user, i.e., the one requesting
media object “9”, and serve it using the free channel. However, if we take a closer look at the
users waiting in the queue, then we can find that there is another user in the third position of the
queue who is also requesting media object “9”. Thus instead of dedicating the free channel to
serve the head-of-line user, we can serve both users simultaneously using multicast streaming.
The two users can then simply join the same multicast group and the network will forward the
media data to both of them.
In this case the system served two users using just one multicast media stream instead of two
unicast media streams. Conceivably, we can achieve better efficiency if there are more users
in the queue and/or there are fewer choices of media objects. Note that to the end user the only
trade-off is the added complexity of supporting multicast in the client application (ignoring
interactive playback control).
Obviously, in practice there may not be many users waiting in the queue when a channel
becomes available. If there is only one queueing user requesting a media object, then multicast
will not offer any advantage over unicast. One way to further improve the batching efficiency
is to artificially delay users to wait for more users to join the batch, even if a free channel is
available.
On the other hand, the system is also not limited to serve the waiting users in a FIFOmanner.
Reconsidering the example in Figure 17.1, we can observe that while there are two queueing