Adaptive Media Streaming - Scalable Continuous Media Streaming Systems - page 151

1000000

900000

800000

700000

600000

500000

400000

300000

200000

100000

0

0 10 20 30 40 50 60 70 80 90 100

Trace No.

AVS (prefetch_known) AVS (prefetch_unknown)

Figure 8.5b Effect of knowledge of prefetch on the transcoded video bit-rate

8.6.2 Effectiveness of Preemptive Rate Control
To investigate how much performance gain can be obtained from preemptive rate control, we
run two sets of simulations for all 94 traffic traces, one with segment-based rate control and
the other with preemptive rate control and plot the results in Figure 8.6.
In all 94 traces, preemptive rate control achieves lower underflow ratios compared to
segment-based rate control. On average, the rebuffering ratio is reduced by 43% when preemp-
tive rate control is used while the average bit-rate is only 2% lower. Nevertheless, preemptive
rate control does require more complex transcoders and thus further investigation is needed to
quantify the gains and the tradeoffs.

8.6.3 Comparison with the CR Algorithm
In this section, we compare the presented rate adaptation algorithm (the AVS algorithm) with
the current state-of-the-art algorithm proposed by Cuetos and Ross [1-2] (the CR algorithm).
In the CR algorithm, there is a control parameter α (0 ≤ α ≤ 1) that can affect the algorithm's

performance. To find the optimal value for , it is necessary to know the network bandwidth

availability over the entire duration of the video session. This is clearly not possible in practice.
Thus, to obtain performance results for the CR algorithm, we ran 2,000 simulations with
the control parameter

α

α varied from 0 to 1 with a step size of 0.0005. We find that the optimal

value for depends heavily on the particular traffic trace chosen, and can range from 0 to 0.25

over the 94 traces. The optimal value for

α

is defined as the value that brings the rebuffering
ratio to within 1% of the minimum and gives that maximum average video bit-rate.
As the optimal

α

is not known a priori , in comparing CR with AVS, we use the rebuffer-
ing ratio and the video bit-rate averaged over all 2,000 simulations, as well as the optimal
results for comparison. The results are shown in Figure 8.7a, 8.7b. We observe that the AVS

α