Title page
Abstract
Contents
I. Introduction 9
II. Related Work 14
III. Policing For Minimum Rate Guarantee 20
3.1 Policing for remaining bandwidth 21
3.2 Policing for insufficient bandwidth 23
IV. Techniques for Minimum Rate Guarantee in TCP-MR 25
4.1 Congestion Avoidance Algorithm 25
4.1.1 Minimum window adjustment 26
4.1.2 Residual window adjustment 32
4.2 Slow start Algorithm 33
4.3 Loss Recovery Algorithm 36
4.4 Guidelines for Parameter Settings 40
4.4.1 Increasing parameter σ 40
4.4.2 Decreasing parameter Υ 41
V. Performance Evaluation 44
5.1 Experimental Results 44
5.1.1 FTP background traffic 45
5.1.2 TCP-MR background traffic 47
5.1.3 Random lossy link 53
5.1.4 UDP background traffic 53
5.1.5 Multiple bottleneck scenario 53
VI. Summary and Conclusions 58
국문요약 59
References 60
Acknowledgement 66
Figure 2.2 : ERED queueing policy 16
Figure 2.1 : General Drop-tail policy 16
Figure 2.3 : Quick-Start vs Slow-Start 18
Figure 4.1 : Bottleneck queue status when it severely occupies by TCP flows 28
Figure 4.2 : Bottleneck queue status when it fully occupies by TCP-MR flows 29
Figure 4.3 : Different loss patterns 30
Figure 4.4 : Window adjustment of elastic and non-elastic flows 36
Figure 4.5 : TCP-MR throughput with Sack option enabled 37
Figure 4.6 : the number of timeouts with Sack option enabled 37
Figure 4.7 : Residual Throughput vs. Different Increasing Parameter(1 TCP-MR flow, 1 TCP flow) 42
Figure 4.8 : Residual Throughput vs. Different Increasing Parameter(1 TCP-MR flow, 7 TCP flow) 42
Figure 5.1 : Single Bottleneck topology 44
Figure 5.2 : Average Throughput vs. Number of background TCP ows 47
Figure 5.3 : Instance Throughput of three TCP-MR flows (minimum rate of 1Mbps) 48
Figure 5.4 : Instance throughput of TCP-MR flows competing with TCP flows : two TCP-MR flows(1Mbps, 200Kbps) start at 0 seconds, five TCP flows start at 30 seconds, five TCP flows start at 60 48
Figure 5.5 : Instance throughput of TCP-MR flows competing with TCP flows : ten TCP flows start at 0 seconds, a TCP-MR flow(1Mbps) starts at 30 seconds, and another TCP-MR flow(200Kbps) starts 49
Figure 5.6 : Single Bottleneck topology with RTT heterogeneity 49
Figure 5.7 : Instance throughput of TCP-MR flows competing with TCP flows : RTT heterogeneity, two TCP-MR flows(1Mbps, 200Kbps) start at 0 seconds, five TCP flows start at 30 seconds, five TCP 50
Figure 5.8 : Instance throughput of TCP-MR flows competing with TCP flows : RTT heterogeneity, ten TCP flows start at 0 seconds, a TCP-MR flow(1Mbps) starts at 30 seconds, and another TCP-MR 50
Figure 5.9 : Average Throughput vs. Number of background TCP-MR flows 51
Figure 5.10 : Instance Throughput of three TCP-MR flows (minimum rate of 1Mbps) 52
Figure 5.11 : Effectiveness of insufficient bandwidth detection algorithm(insufficient bandwidth because of newly joined TCP-MR flows) 55
Figure 5.12 : Effectiveness of insufficient bandwidth detection algorithm(join into the fully utilized network) 56
Figure 5.13 : Average Throughput vs. Loss rate 57
Figure 5.14 : Average Throughput vs. Number of background UDP flows 57