A Concurrent-Push Parallel Server Architecture - Scalable Continuous Media Streaming Systems - page 180

10.3.5 System Response Time
Response time is defined as the time from the user request for a new video session to the time
actual video playback starts. This delay comprises two components: scheduling delay , and
prefill delay . Scheduling delay is the time from a client sending a new-session request to the
time transmission starts at the server. It is easy to see that the worst-case scheduling delay is
two service rounds (see Figure 10.2):

2 N S Q
R V

D S = (10.21)

Prefill delay is the time from the server starts transmission to the time the first y groups of
client buffers are fully filled with data. Using equation (10.10), the worst-case prefill delay can
be obtained from

D P = max { F ( y − 1) }− t 0 (10.22)

or

f + = f +

D P = yT F + max { δ }+ ( y + 1) T F + (10.23)

3 f + − T F +

f − − T E

f + = +

T F

10.4 Asynchronous Grouped Sweeping Scheme
The results in the previous section reveal an important characteristic of the concurrent push
algorithm, namely, the server buffer requirement, the client buffer requirement, and the response
time all increase with the number of servers in the system. Therefore, the scalability of the
system will either be limited by the economy of memory buffers or the tolerance of the system
response time by the user. In this section, we propose an extension of the Grouped Sweeping
Scheme (GSS) [8], called Asynchronous Group Sweeping Scheme (AGSS) to substantially
reduce server buffer requirement, and scheduling delay.

10.4.1 Extending the Grouped Sweeping Scheme
The original GSS algorithm [8] is designed for scheduling retrieval requests in a magnetic disk.
The traditional First-In-First-Out scheduling algorithm has poor disk utilization in continuous-
media applications because in the worst case the disk arm may need to seek back and forth
between the innermost track and the outermost track, thus wasting a lot of time in seeking
(cf. Chapter 3). Instead, some systems use the SCAN scheduling algorithm to reduce seek-
time overhead by serving requests while the disk arm scans across the disk surface. However,
this approach requires two buffers per stream because requests may be served out of order and
in the worst case, two requests for the same stream may be served in a back-to-back manner.
The GSS algorithm is designed to strike balance between minimizing seek-time overhead
and minimizing buffer requirement by serving streams in groups. Streams within a group are
served using SCAN to reduce seek-time overhead while the groups are served in a fixed order to