Adaptive Media Streaming - Scalable Continuous Media Streaming Systems - page 149

8.6 Performance Evaluation
In this section, we use a trace-driven simulator written in ns-2 [12] to evaluate the performance
of the presented adaptation algorithm (denoted by AVS) and compare it with the current state-
of-the-art algorithm proposed by Cuetos and Ross [1-2] (denoted by CR).
Figure 8.4 depicts the simulated network topology. We use the common NewReno TCP [13-
14] as the transport protocol to deliver the video data to the client. Cross-traffic is generated
from a packet trace file obtained from Bell Labs [15-16]. 1 The trace file captured 94 hours of
network traffic passing through a firewall. We divide the 94-hour trace file into 94 1-hour trace
files and run a simulation for each 1-hour trace file to evaluate the algorithms' performance
under different cross-traffic scenarios.
Both the streaming traffic and the cross-traffic share a link of R Mbps as shown in Fig-
ure 8.4. For each simulation, we adjust R so that the network has just sufficient bandwidth to
stream the video, i.e., R

c , where c is the average data rate of the cross traffic. We
summarize the system settings in Table 8.1.
We use two performance metrics, namely, rebuffering ratio and average video bit-rate, to
evaluate the algorithms' performance. Rebuffering ratio is defined as the proportion of frames

=

r max +

Streaming Client

Streaming Server

Video Stream

R Mbps, 100ms

Cross -Traffic

Avg_rate = c

Cross-Traffic Generator Cross-Traffic Receiver

Figure 8.4 Network topology used in simulations

Table 8.1 System parameters used in simulations

Parameters Value

Prefetch duration 5 seconds

M 1 second

r max 1.1 Mbps

r min 200 kbps

Video Length 3000 seconds

TCP MSS 1500 bytes