A Hybrid Architecture - Scalable Continuous Media Streaming Systems - page 351

20

18

16

14

12

10

8

6

Analytical Results
Simulation Results
Experimental Results

4

2

0

0 1 2 3 4 5

Arrival Rate (requests/second)

Figure 19.16 Comparison of latencies obtained from analysis, simulation and benchmarking

With the SS-VoD prototype, we conducted extensive experiments to obtain measured bench-
mark results to verify against the analytical and simulation results. We developed a traffic
generator in order to simulate a large number of client requests. The service node runs on
a Compaq Proliant DL360 serving one movie of length 120 minutes with 30 channels, each
at 1.5Mbps. The clients are ordinary PCs and all machines are connected using a layer-3
IP switch with hardware IP multicast support. We measured the start-up latency for arrival
rates ranging from 1 to 5 requests per second. Each benchmark test runs for a total of six
hours. Benchmark data collected during the first hour is discarded to reduce initial condition
effect.
Figure 19.16 compares the start-up latencies obtained from analysis, simulation, and bench-
marking respectively. We observe that the benchmarking results agree very well with the
analytical results and simulation results. Note that the latencies obtained from benchmarking
are consistently larger than those obtained from simulation. We believe that this is due to the
non-zero processing delay and network delay in the system, both of which have been ignored
in the simulation model.

19.6 Summary
In this chapter, we investigated a Super-Scalar Video-on-Demand (SS-VoD) architecture that
can achieve super-linear scalability by integrating techniques of batching, patching, and pe-
riodic broadcasting. In designing the SS-VoD architecture the focus is on its practicality and
the implementation and deployment complexities. For example, instead of adopting more so-
phisticated open-loop algorithms to schedule the static multicast channels, we employed the
simple staggered periodic multicast schedule that enables us to implement interactive play-
back control such as pause-resume, slow motion, and seeking in a simple yet efficient way.
Moreover, the staggered schedule also requires significantly lower client buffer requirement
and more importantly, eliminates the need to switch multicast channels during a video session.