performance is lowered by about 65%(compare stretch= 1 we study if an optimal static allocation of non-overlapping ith stretch 10)due to interference between aggressive channels across APs could eliminate interference altogether Http flows(s=5s). The same is true for the performance Second we present a preliminary investigation of the effect of the Ftp flow in Figure 4(b). For a less aggressive Http of reducing the transmit power levels at Aps on the interfer component(s= 20s)the performance of the Http flows is ence experienced We also investigate how transmit power 0% inferior while the Ftp flow suffers by about 36% control improves the total capacity of a chaotic, network, as well as the fairness in the allocation of the capacity among individual APs 0.09 4.3.1 Effect of Optimal Static Channel Allocation 08 0.07 We performed simulations on the 20-node topology of Fig 06 ure 2(a), where the APs are statically assigned one of the 0.05 three non-overlapping channels(1, 6 and 11)such that no two neighboring APs share a channel, whenever possible. As 003圈哲普吾翻善—每…量: an illustration, Figure 2(b) shows the lay-out of APs that were all assigned channel 1 by this scheme. Stretch (a)httpperformance 5S, W: comb ftp ML: comb ft3“母 0.8 日善醫善…:皇…量… 0.01 5 04 0*术 Stretch VL: comb ftp2 (a)httpperformance 0. pmb ftpa 0s,WL: comb ftp3…母 567 Stretch 0.8 (b)FTP 0.7 0.6 Figure 5: Average performance of Http and ftP 04 Ing (i. s: 5S, WL: comb ftp2 the comb-ftp 2, 3 workloads) and for D=3 0.1 203WCcm出:3- Impact of traffic volume. Figures 5(a) and(b) show 12345678910 the average performance of the Http and Ftp flows re spectively, in simulations with a few more competing FTP Stretch Hows--i.e, the comb-ftp2 and comb-ftps workloads--for (b)FTP performance D=3. The performance impact on Http and Ftp flows hecaseswheretheFigure6:PerformanceofhttpandFtpflowswith ITTP component of these workloads is not very aggressive see the curves corresponding to s= 20s in Figure 5(b)) optimal static assignment of APs to the ree non- overlapping channels. The transmit power level is Using realistic channel assignments. We also performed set at 15dBm, corresponding to a reception range of simulations on the 20-node topology, where the APs were 31m tatically assigned channels based on the distribution in Ta ble 3. However. we note similar levels of interference and The performance of htTp and Ftp flows in these simu pact on performance as observed above. This is because lations are shown in Figure 6(a)and(b), respectively. The more than half the APs in this simulation were assigned average performance of both htTp and Ftp flows improves channel 6, which was the most predominant channel er ployed by most APs according to our measurements significantly. Comparing with Figures 5(a) and(b)respec- tively, we note that the performance curves flatten out earlier on account on the sparse nature of the interference 4.3 Limiting the Impact of Interference graph. Nevertheless, the impact of interference can still be Inthissectionweexploretheeffectoftwosimplemechaseentheaveragehttpperformanceisabout25%inferior nisms on mitigating interference in chaotic networks: First at stretch 1 compared to the case when no nodes interperformance is lowered by about 65% (compare stretch = 1 with stretch = 10) due to interference between aggressive HTTP flows (s = 5s). The same is true for the performance of the FTP flow in Figure 4(b). For a less aggressive HTTP component (s = 20s) the performance of the HTTP flows is 20% inferior, while the FTP flow suffers by about 36%. 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 1 2 3 4 5 6 7 8 9 10 Normalized HTTP Performance Stretch s: 5S, WL: comb ftp2 s: 5S, WL: comb ftp3 s: 20S, WL: comb ftp2 s: 20S, WL: comb ftp3 (a) HTTP performance 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 8 9 10 Normalized FTP Performance Stretch s: 5S, WL: comb ftp2 s: 5S, WL: comb ftp3 s: 20S, WL: comb ftp2 s: 20S, WL: comb ftp3 (b) FTP performance Figure 5: Average performance of HTTP and FTP flows at higher competing FTP traffic levels (i.e., the comb-ftp{2, 3} workloads) and for D = 3. Impact of traffic volume. Figures 5(a) and (b) show the average performance of the HTTP and FTP flows, re￾spectively, in simulations with a few more competing FTP flows—i.e., the comb-ftp2 and comb-ftp3 workloads—for D = 3. The performance impact on HTTP and FTP flows is slightly more pronounced, even for the cases where the HTTP component of these workloads is not very aggressive (see the curves corresponding to s = 20s in Figure 5(b)). Using realistic channel assignments. We also performed simulations on the 20-node topology, where the APs were statically assigned channels based on the distribution in Ta￾ble 3. However, we note similar levels of interference and impact on performance as observed above. This is because more than half the APs in this simulation were assigned channel 6, which was the most predominant channel em￾ployed by most APs according to our measurements. 4.3 Limiting the Impact of Interference In this section, we explore the effect of two simple mecha￾nisms on mitigating interference in chaotic networks: First, we study if an optimal static allocation of non-overlapping channels across APs could eliminate interference altogether. Second, we present a preliminary investigation of the effect of reducing the transmit power levels at APs on the interfer￾ence experienced. We also investigate how transmit power control improves the total capacity of a chaotic, network, as well as the fairness in the allocation of the capacity among individual APs. 4.3.1 Effect of Optimal Static Channel Allocation We performed simulations on the 20-node topology of Fig￾ure 2(a), where the APs are statically assigned one of the three non-overlapping channels (1, 6 and 11) such that no two neighboring APs share a channel, whenever possible. As an illustration, Figure 2(b) shows the lay-out of APs that were all assigned channel 1 by this scheme. 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 1 2 3 4 5 6 7 8 9 10 Normalized HTTP Performance Stretch s: 5S, WL: comb ftp2 s: 5S, WL: comb ftp3 s: 20S, WL: comb ftp2 s: 20S, WL: comb ftp3 (a) HTTP performance 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1 2 3 4 5 6 7 8 9 10 Normalized FTP Performance Stretch s: 5S, WL: comb ftp2 s: 5S, WL: comb ftp3 s: 20S, WL: comb ftp2 s: 20S, WL: comb ftp3 (b) FTP performance Figure 6: Performance of HTTP and FTP flows with optimal static assignment of APs to the three non￾overlapping channels. The transmit power level is set at 15dBm, corresponding to a reception range of 31m. The performance of HTTP and FTP flows in these simu￾lations are shown in Figure 6(a) and (b), respectively. The average performance of both HTTP and FTP flows improves significantly. Comparing with Figures 5(a) and (b) respec￾tively, we note that the performance curves “flatten out” earlier on account on the sparse nature of the interference graph. Nevertheless, the impact of interference can still be seen: the average HTTP performance is about 25% inferior at stretch = 1 compared to the case when no nodes inter-