Fig. 1. Samples of connections evolution within a simplified TCP OMQN model.

Fig. 2. The OMQN model of TCP NewReno.

Fig. 3. Sketch of the recursive procedure to derive the probability mass function of .

Fig. 4. Impact of R on the distribution of the TCP connection completion time.

Fig. 5. Single bottlenecks scenario with homogeneous connections.

Fig. 6. CDF and CCDF of the completion time. Single bottleneck, homogeneous connections with N_s=100 and ρ_n=0.9; Analysis ( ms and P_L=1.64%) and simulation ( ms and P_L=1.61%).

Fig. 7. Analysis and simulation. Probability density function. Single bottleneck, homogeneous connections with N_s=100 and ρ_n=0.9.

Fig. 8. CCDF, single bottleneck, connections with N_s=10 and ρ_n=0.95. Analysis ( ms and P_L=0.6%) and simulation ( ms and P_L=0.5%).

Fig. 9. Two bottlenecks scenario, non-homogeneous connections. Setup (left) and simulation results (right). Flows from router 2 to router 1 ( ms and P_L=0.7%) and from router 3 to router 1 ( ms and P_L=2.7%).

Fig. 10. Analysis and simulation. Complement of the cumulative distribution function. Two bottlenecks scenario, non-homogeneous connections. Connection length N_s=100 (left) and N_s=10 (right).

Fig. 11. CDF and CCDF. Single bottleneck scenario, mixed UDP and TCP traffic. Connection length N_s=10. Analysis and simulation ( ms and P_L=29%).

Fig. 12. Analytical results. Mean of the completion time as a function of the connection length and of the segment loss probability.

Fig. 13. Analytical results: 0.95 (left) and 0.99 (right) completion time quantiles as a function of the connection length N_s and the segment loss probability P_L.