A Hybrid Architecture - Scalable Continuous Media Streaming Systems - page 335

tuple: ), is the length

of service for each video title being served. Therefore, each service node will have M such
admission tuples and each admission controller will have MK such admission tuples, where K
is the total number of service nodes in the system. Both the counter and the length fields are
initially set to zero.
Now with the admission tuples, the admission procedure proceeds as follows. For requests
that cannot be statically-admitted, the admission controller will first check the counter in the
admission tuple for the requested video title. If the counter A C is zero, then the counter is
incremented by 1, and the length field is set according to

A C , A L } , where A C ={ 0 , 1 } is the counter, and A L ,0 ≤ A L ≤ ( T R − 2 δ

{

A L = a i − t m (19.3)

which is the length of time passed since the beginning of the last multicast. In other words, this
particular client will occupy the dynamic channel for a duration of A L seconds for patching
purpose. At the same time a START request carrying the requested video title and the length
field A L will be sent to a service node and the admission controller enters the STARTED state.
If another request for the same video title arrives during the STARTED state, say, at time a i + 1 ,
the admission controller will not send another request to the service node, but will just update
the local length field according to

A L = a i + 1 − t m (19.4)

This process repeats for all subsequent requests arrived during the STARTED state. As a result,
only one START request will be sent to the service node regardless of how many requests are
received during the STARTED state, thereby significantly reducing the processing overhead at
the service node.
At the service node side, upon receiving a START request from the admission controller,
the service node will wait for a free channel from the N D dynamic multicast channels to start
transmitting the video title for a duration of A L seconds as shown in Figure 19.4. Once a
channel becomes available, a START reply will be sent back to all admission controllers to
announce the commencement of the new transmission.
The admission controllers, upon receiving the START reply, will do one of two things. If
the local counter value is 1, then both the counter and the length fields are zeroed and the
admission process is completed. If the counter is greater than 1, then the admission controller
will send an EXTEND request to the service node to extend the transmission duration according
to the value of the local length field A L . Note that in this case, the length field at the admission
controller will be larger than the length field at the service node because only the length field at
the admission controller is updated for subsequent requests for the same video title. The length
field at the service node is always the one for the first request. After receiving EXTEND requests
from the admission controllers, the service node will update the transmission duration to the
largest one among all EXTEND requests. Transmission will stop after the specified transmission
duration expires. Note that the service node does not need to wait for any EXTEND request to
begin streaming. Streaming will begin as soon as a free dynamic channel becomes available.
The purpose of the EXTEND request is to increase the transmission time of the dynamic channel
to cater for subsequent requests in the same batch that require a longer patching duration.