Throughput Improvement

Improving the throughput of a connection when the congestion occurs in the back-ward path is one of the design goals of RoVegas. In this subsection, we investigate the throughputs of Vegas and RoVegas in two types of backward congestion. One is the congestion caused by network asymmetry, the other is the congestion caused by additional backward traffic.

The first network topology for the simulations is shown in Fig. 3.3. Sources, destinations, and routers are expressed as Si, Di, and Ri respectively. A source and a destination with the same suffix value represent a traffic pair. The bandwidth and propagation delay are 10 Mb/s and 1 ms for each full-duplex access link, 1.6 Mb/s and 20 ms for the connection link from R1 to R2, and Cb and 20 ms for the connection link from R2 to R1, respectively. Cb is set based on the normalized asymmetric factor k. For example, if k = 4 and the size of data packet and ACK are 1 Kbytes and 40 bytes respectively, then C is set to 16 Kb/s.

0

Figure 3.4: Throughput of Vegas in asymmetric networks.

Asymmetric Networks: To evaluate the throughputs of Vegas and RoVegas in asymmetric networks, different values of k are used. A source S1 of either Vegas or RoVegas sends data packet to its destination D1. The size of each FIFO queue used in routers is 10 packets. Figures 3.4 and 3.5 exhibit the throughput performance of Vegas and RoVegas in asymmetric networks respectively.

By observing the results shown in Fig. 3.4, with the increasing value of k from 2 to 32, the throughput of Vegas degrades accordingly. As our analysis depicted in Eq. (3.17), the throughput of Vegas in this scenario should be uf/k (data packets per second). Obviously, the simulation results conform to our previous analysis.

Comparing the results of Fig. 3.5 with that of Fig. 3.4, we can find that the throughput of RoVegas is much greater than that of Vegas. With k = 2, RoVegas maintains a high throughput at 1587.2 Kb/s in which the backward congestion seems not existing. The throughput ratios of RoVegas to Vegas in steady state are about 2 and 3 for k=2, 4 and k=8, 16, 32 respectively. Notably, all the simulation results shown in Fig. 3.4 and Fig. 3.5 are consistent with our previous analysis.

Symmetric Network With Backward Traffic: Asymmetric networks should not be the only reason that causes backward congestion. Actually, even in a

sym-0

Figure 3.5: Throughput of RoVegas in asymmetric networks.

metric network the backward congestion may still occur. We use a VBR source with 1.44 Mb/s averaged sending rate to examine the throughputs of Vegas and RoVegas separately in a single bottleneck network as shown in Fig. 3.3. The capacity of the backward bottleneck, Cb, is set to 1.6 Mb/s. A source of either Vegas or RoVegas is attached to S1 and a VBR source is attached to S2. The S1 starts sending data at 0 second, while S2 starts at 50 second. Figure 3.6 depicts the throughput comparison between Vegas and RoVegas.

As shown in Fig. 3.6, when the traffic source is Vegas only (0–50 second), it achieves high throughput and stabilizes at 1.6 Mb/s. However, the performance of Vegas degrades dramatically as the VBR source starts sending data. Although the overhead induced by AQT option slightly lower the throughput of RoVegas (0.8 %) during the preceding 50 seconds, nevertheless, RoVegas maintains a much higher throughput than that of Vegas while the backward congestion occurs. With the inference of the backward VBR traffic, the average throughput of Vegas is 521 Kb/s and RoVegas is 1092 Kb/s. Since we use the same traffic pattern of the VBR source while the throughput of Vegas or RoVegas is examined. Thus there are some synchronized throughput fluctuations between Vegas and RoVegas.

0 200 400 600 800 1000 1200 1400 1600 1800

0 50 100 150 200

Time (s)

Throughput (Kb/s)

Vegas RoVegas

Figure 3.6: Throughput comparison between Vegas and RoVegas with the backward traffic load is 0.9 in the single bottleneck network topology.

To evaluate the average throughputs of Vegas and RoVegas with different back-ward traffic loads, we set the VBR traffic loads to vary from 0 to 1. The traffic sources are the same as the above descriptions but the sources of either Vegas or RoVegas and VBR start at 0 second. The simulation period is 200 seconds for each sample point. From the simulation results shown in Fig. 3.7, we can find that when the backward traffic load is not zero, RoVegas always achieves a higher average throughput than Vegas. For example, as the backward traffic load is 1, RoVegas achieves a 4.1 times higher average throughput in comparison with that of Vegas.

In the parking lot configuration as shown in Fig. 3.8, we use three VBR sources each with 1.28 Mb/s averaged sending rate to examine the throughputs of Vegas and RoVegas. The bandwidth and propagation delay of each full-duplex access link and connection link are 10 Mb/s, 1 ms and 1.6 Mb/s, 10 ms respectively. The source of either Vegas or RoVegas are attached to S1, and three VBRs are attached to S2

to S4 respectively. The TCP source from either Vegas or RoVegas starts sending data at 0 second, and then three VBR sources from S2 to S4 successively enter the network every 100 seconds.

0 Traffic Load of VBR Source

Avergae Throughput (Kb/s)

Vegas RoVegas

Figure 3.7: Average throughput versus different backward traffic loads for Vegas and RoVegas in the single bottleneck network topology.

S1

Figure 3.8: A parking lot network topology for investigating throughputs of Vegas and RoVegas when the congestion occurs on the backward path.

From the simulation results presented in Fig. 3.9 we can observe that when the traffic source is TCP only (0–100 second) both Vegas and RoVegas could fully utilize the bandwidth (due to the overhead induced by AQT option, the throughput of RoVegas is slightly lower than Vegas). However, as the VBR sources successively enter the network, Vegas suffers a serious throughput reduction. Under the same environment, RoVegas features a much better throughput performance compared with that of Vegas. The average throughput ratio of RoVegas to Vegas during 100–200, 200–300, and 300–400 second are 2.00, 2.77, and 3.46 respectively.

0 200 400 600 800 1000 1200 1400 1600 1800

0 50 100 150 200 250 300 350 400

Time (s)

Throughput (Kb/s)

Vegas RoVegas

Figure 3.9: Throughput comparison between Vegas and RoVegas with the backward traffic load is 0.8 in the parking lot network topology.

loads in the parking lot network are also examined. The traffic sources of either Vegas or RoVegas and three VBRs start at 0 second. The VBR traffic loads vary from 0 to 1 accordingly. From the simulation results shown in Fig. 3.10 we can find that as the backward traffic load is not zero, RoVegas always achieves a higher average throughput than Vegas, especially when the backward traffic load is heavy.

For example, as the backward traffic load is 1, the average throughput ratio of RoVegas to Vegas is 14.09.

Obviously, we have demonstrated that RoVegas significantly improves the con-nection throughput when the backward path is congested.