Information Technology Reference
In-Depth Information
100
80
60
40
20
0
0
0.2
0.4
0.6
0.8
1
Arrival Rate (requests/second)
SS-VoD
UVoD
Figure 19.11
Comparison of optimal channel allocation in SS-VoD and UVoD
the unicast channel to prevent operating the unicast channels near full utilization. By contrast,
the same situation does not occur in SS-VoD because a dynamic multicast channel can batch
and serve multiple waiting requests. Moreover, the batching efficiency increases for longer
waiting time, thus compensating for the increases in the arrival rate. This remarkable property
of SS-VoD greatly simplifies system deployment as one will not need to reconfigure the system
with a different channel partition policy if the user demand changes.
19.4.3 Latency Comparisons
Figure 19.12 plots the latency for SS-VoD, UVoD, TVoD, and NVoD for arrival rates up to
5 requests per second. The service node (or video server for TVoD/NVoD) has 50 channels and
serves a singlemovie of length 120minutes. The first observation is that except forNVoD, which
has a constant latency of 72 seconds, the latency generally increases with higher arrival rates as
expected. For TVoD, the server overloads for arrival rates larger than 1
10
−
4
requests per
second. UVoD performs significantly better with the latency asymptotically approaches that
of NVoD. SS-VoD performs even better than UVoD, and the latency levels off and approaches
5.6 seconds, or a 92% reduction compared to UVoD.
It is also worth noting that the performance gain of SS-VoD over UVoD does not incur any
trade-off at the client side. Specifically, the buffer requirement and bandwidth requirement are
the same for both SS-VoD and UVoD. The only differences are the replacement of the dynamic
unicast channels in UVoD with dynamic multicast channels in SS-VoD, and the addition of the
more complex admission procedure in the admission controller.
.
16
×
Search WWH ::
Custom Search