Information Technology Reference
In-Depth Information
Server
Client
Server
Client
Start new video
Start new video
Video data
Request
Video data
Video data
.
Video data
Video data
Video data
Request
Video data
Video data
Video data
Video data
.
Video data
.
(
a
)
(
b
)
Server Push
Client Pull
Figure 9.7
Two approaches to data delivery: server push and client pull
receiving the request, the server will retrieve the data from the disk and send it back to the
client. This approach is called
client pull
for obvious reasons.
9.4.2 Inter-Server Synchronization
Most of the studies on single-server VoD systems employ the server-push delivery model.
This model allows the system designer to devise periodic schedules at the server for reading
data off the disk and then transmitting to the client. If we extend the server-push model to
a parallel video server, a new problem arises due to the parallel transmissions from multiple
independently running servers.
For example, let us consider a system with
N
S
servers using fixed-size space striping. To
start a new video session, a client will send a request to the proxy, which in turn will send
requests to all
N
S
servers to start a new video session. Due to variations in processing delay,
network delay, and scheduling delay, the servers will start transmitting data at different time
instances. It is possible that the first stripe unit will arrive at the proxy later than the subsequent
stripe units. Consequently, the proxy has to buffer the later stripe units to wait for the first
stripe unit to arrive for playback, thereby increasing the client buffer requirement and start-up
delay.
This synchronization problem has been investigated by Biersack
et al.
in [8]. For scenarios
where network delays between servers and a client are different, they proposed adding different
delays to the starting times of each server to compensate for delay differences. They also
extended this model to include bounded network delay jitters. In another study by Buddhikot
and Parulkar [11], they designed a closely-coupled system in which each storage node is
connected by a custom high-speed interconnection network (APIC). The proximity of the
storage nodes enables them to be accurately synchronized through the common APIC. We will
return to this synchronization issue in subsequent chapters.
Search WWH ::
Custom Search