A Hybrid Architecture - Scalable Continuous Media Streaming Systems - page 340

19.2.3 Seeking
Seeking is the change from one playback point to another. Typically, the user initiates seeking
either by giving a new destination time offset or by means of using a graphical user interface
such as a slider or a scroll bar. SS-VoD can support different types of seeking depending
on the seek direction, seek distance, and the state of the client buffer and static multicast
channels. Specifically, due to patching, the client buffer typically has some advance data cached.
Moreover, some past video data will also remain in the client buffer until being overwritten
with new data. Hence, if the new seek position is within the range of video data in the client
buffer, seeking can be implemented simply by changing the playback point internally.
Now if the seek position, denoted by t s , lies outside the client buffer, then the client may
need to switch multicast channels to accomplish the seek. Let t i ,

1 be the
current playback points of the N S static multicast channels and assume the client is currently
on channel x . Then the client will choose the nearest channel to restart playback by finding
the channel j so that the seek error

i = 0 , 1 ,..., N S −

ε = {|

t j − t s | , | t s − t j |} is minimized. Note that the

current channel may happen to be the nearest channel and in this case, the client simply seeks
the oldest data in the buffer if t s is earlier than the current playback point, or seeks the newest
data in the buffer otherwise.
Clearly in the previous case the seek operation may not end up in the precise location
specified by the user and the seek error can be up to T R /

min

2 seconds. In return, this seeking
algorithm can be supported without incurring server overhead or additional client buffer. If
more precise seeking is needed, then one will need to make use of a dynamic multicast channel
to merge the client back into an existing static multicast channel. Further research will be
needed to develop efficient yet precise seeking algorithms.

19.3 Performance Modeling
In this section we present an approximate performance model for the SS-VoD architecture.
While an exact analytical solution does not appear to be tractable, we were able to derive an
approximate model that can be solved numerically. The purpose of this performance model is to
assist system designers to quickly evaluate various design options and to perform preliminary
system dimensioning. Once the approximate system parameters are known, one can turn to a
more detailed simulation to obtain more accurate performance results.
The primary performance metric we use in this study is start-up latency, defined as the time
fromwhen a client submits a request to the admission controller to the time when the beginning
of the requested video starts streaming. For simplicity, we assume there is a single video title
stored in a service node and ignore network delay, transmission loss, and processing time at
the admission controller.
In the following sections, we will first derive the average waiting time for statically-admitted
clients and dynamically-admitted clients, and then investigate the configuration of the admis-
sion threshold and the channel partitioning policy. We will compare results computed using
this approximate performance model with the simulation results in Section 19.4.1.

19.3.1 Waiting Time for Statically-Admitted Clients
As described in Section 19.1.2, there are two ways to admit a client into the system. The first
way is admission through a static multicast channel as shown in Figure 19.5. Given that any