Closed-Loop Algorithms - Scalable Continuous Media Streaming Systems - page 295

17

Closed-Loop Algorithms

This chapter presents closed-loop algorithms to support on-demand multicast streaming.
In closed-loop algorithms, the transmission schedule, reception schedule, and/or playback
schedule are dynamically adjusted according to the current workload, i.e., number of active
media streams, their playback points, etc. The objective is to merge as many clients onto as
fewmulticast data streams as possible to reduce resource utilization. This chapter illustrates
this closed-loop approach by describing the techniques of batching, patching, prefix caching,
and piggybacking. These techniques not only can be applied individually, but also can be
combined to achieve even more performance gains.

17.1 Introduction
In this chapter we present four techniques to improve the efficiency of using multicast to
support on-demand media streaming. Compared to media broadcasting, on-demand media
streaming allows the users to begin viewing a selected content at their chosen time. This model
is applicable to many applications, such as video-on-demand, online education, digital library,
and so on. Supporting arbitrary start times for different users, however, conflicts with the
constraint of network multicast, where all users of the same multicast group will receive the
same set of media data.
To resolve this conflict, we will need to accept trade-offs in other dimensions. The following
sections describe four techniques to improve multicast efficiency by trade-offs in different
dimensions. These four techniques are chosen to illustrate the possibilities of different trade-
offs and the resultant performance gains, and as such are not meant to be exhaustive. In
fact, there are many other sophisticated closed-loop algorithms in the literature which are not
covered in this chapter. We will briefly discuss some of them in Section 17.6 and interested
readers are referred to the literature for more details.

17.2 Batching
Batching is a general technique that has been applied to many engineering problems. The
principle is to group together similar tasks so that they can be processed in a more efficient