Overview of Multicast Streaming - Scalable Continuous Media Streaming Systems - page 293

. . .

Multicast Ch x : A B C

time

. . .

Client i Playback:

A C

Client performs a jump to playback segment “C”

Client begins playback

Figure 16.6 A client changing the playback point can no longer receive data from the same multicast
channel

data to this user using a separate data stream but then the new data stream is not shared with
other users, thus defeating the efficiency gains of using multicast in the first place. This is still
an open problem that warrants more research.

16.4 Techniques for On-Demand Multicast Streaming
To support on-demand multicast streaming, it is necessary to trade off other dimensions such as
time, space, and quality to permitmultiple clients to share the samemulticast data transmissions.
The principles are to schedule the transmission, reception, and playback of media data to
enhance sharing of multicast data among multiple clients. These techniques can be broadly
classified into three categories: closed-loop, open-loop, and hybrid algorithms.
In closed-loop algorithms, the multicast media streams are dynamically scheduled according
to the user arrival pattern to merge them into as few shared multicast data streams as possible.
An important characteristic of closed-loop algorithm is that the transmission schedule may
be modified as new clients arrive and as existing clients depart to exploit any opportunities to
merge more clients into fewer multicast channels.
By contrast, in open-loop algorithms (also known as periodic broadcasting), all multicast
transmissions are prescheduled in a fixed pattern irrespective of the user arrival pattern. Thus,
the bandwidth requirement at the media server is fixed regardless of the number of users in the
system. In other words, open-loop algorithms in principle can serve as many concurrent users
as the network will allow. Unlike closed-loop algorithms where transmission schedules are
dynamically determined at the media server (or a separate scheduler), in open-loop algorithms
the complexity is shifted to the clients, who will need to determine which media data and
when to receive media data from one or more multicast channels to sustain continuous media
playback.
Comparing closed-loop and open-loop architectures, the performance (e.g., start-up latency)
of closed-loop architectures depends on the system load (i.e., user arrival rate), and generally
the performance deteriorates with higher system load. By contrast, open-loop architectures
have invariant performance irrespective of the system load. Consequently, at light system
load closed-loop architectures can achieve better performance while open-loop architectures
perform better at high system load.