A Hybrid Architecture - Scalable Continuous Media Streaming Systems - page 345

network where the network routers will need to duplicate and forward the multicast video data
to multiple recipients. Nevertheless, this additional cost is not present at the server (e.g., using
IP multicast) and at the backbone network before fanning out to the access sub-networks and
therefore will be ignored in this study.

19.4.1 Model Validation
To verify accuracy of the performancemodel derived in Section 19.3, we developed a simulation
program using CNCL [4] to obtain simulation results for comparison. A set of simulations
is run to obtain the latency over a range of arrival rates. Each run simulated a duration of
1,440 hours (60 days), with the first 24 hours of data skipped to reduce initial condition effects.
There is one movie in the system, with a length of 120 minutes. We divide available multicast
channels equally into static-multicast and dynamic-multicast channels. We do not simulate
user interactions and assume all users play back the entire movie from start to finish.
Figure 19.9 shows the latency versus arrival rate ranging from 1 10 − 3 to 5.0 requests per

second. We observe that the analytical results are reasonable approximations for the simulation
results. At high arrival rates (e.g., over 1 request per second), the analytical results over-estimate
the latency by up to 5%. As discussed in the beginning of Section 19.3, the analytical model is
primarily used for preliminary system dimensioning. Detailed simulation, while lengthy (e.g.,
hours), is still required to obtain accurate performance results.

×

40

30

20

10

0

1 2 3 4 5

Arrival Rate (requests/second)

Simulation Result (20 channels)
Analytic Result (20 channels)
Simulation Result (30 channels)
Analytic Result (30 channels)
Simulation Result (50 channels)
Analytic Result (50 channels)

Figure 19.9 Comparison of latency obtained from analysis and simulation