Closed-Loop Algorithms - Scalable Continuous Media Streaming Systems - page 299

begins caching media data from the on-going multicast stream, and waits for the system to
allocate a patching stream. The patching stream, once available, will transmit those media data
that the client has missed from the on-going multicast stream, i.e., the “A” segment. Using the
media data received from the patching stream the client then begins playback.
Eventually, the patching stream will complete the transmission of the 'A' segment, which
includes media data from the beginning to playback time ( t a − t 0 ). After that, the client can

continue playback using the cached data and data received from the on-going multicast stream.
Thus the client is effectively merged to the on-going multicast stream. Compared to starting a
new media stream, the cost is reduced from a full stream of L seconds - length of the media
content, to the patching stream of ( t a −

t 0 ) seconds.
Note that patching is orthogonal to batching and thus they can be used together to achieve
even better efficiency. For example, if more clients arrive during the waiting time for an
available patching stream (i.e., within [ t a , t p ] in Figure 17.4), then they can be served using the
same patching stream as shown in Figure 17.5. Two clients arrive at time t a and t b respectively,
missing the 'A' and 'AB' portions of the on-going multicast stream respectively. The clients
are batched together to use the same patching stream to begin playback. Note that in this case
the patching stream will need to be extended to transmit segment 'B' required by client 2 in
addition to segment 'A' required by both clients. In general, the duration of the patching stream
will be determined by the last client arriving before the patching stream begins.
Patching has two trade-offs. First, during the patching duration the client will need to receive
two media streams simultaneously. This obviously doubles the client's bandwidth requirement.
Second, the client will need additional buffers to cache the data received from the on-going

t 0 t a t b t p t q

On-going multicast stream: . . .

A B C

Patching stream:

A B The patching stream is extended for Client 2.

w

time

. . .

Client 1 Playback:

A B C

. . .

Client 2 Playback:

A B C

t 0
t a
t b
w
t p
t q

: Time when the multicast media stream begins.

: Client 1 arrival time
: Client 2 arrival time
: Waiting time for an available patching stream.
: Time when a patching stream is started, also playback begins.
: Time when the patching stream is terminated ( t q = t p + w )

Figure 17.5 Combining patching with batching to further improve efficiency