Fig. 1. Evolution of the throughput of a non-persistent flow with time. The exponential slow start phase is in red. The jump approximation is in black. The file sizes F_i are i.i.d. and sampled according to the distribution function G; the think times T_i are also i.i.d. and are sampled according to F. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2. Dependence of the mean throughput (y-axis) as given by Formula (5) on the loss probability p (x-axis). Here R=0.1 s, 1/β=2 s, 1/μ=2000 pkts. p varies and H=δ(100) (the delta function at 100 pkts).

Fig. 3. Dependence of the mean throughput (y-axis) as given by Formula (5) on H. H=δ(η) and η varies (x-axis). Here R=0.1 s, 1/β=2 s, 1/μ=2000 pkts. where η varies.

Fig. 4. Mean time to transfer a file (y-axis) for H=δ(0). Here R=0.1 s, 1/β=2 s, p=1% when 1/μ varies (x-axis).

Fig. 5. Mean time to transfer a file (y-axis) for H=δ(100) as a function of 1/μ (x-axis). Same parameters as in Fig. 4.

Fig. 6. Comparison of (16) and (10). Here R=0.1 s, 1/β=2, p=1%, q_i=Ai⁻⁴.

Fig. 7. Difference of (16) and (10). Here R=0.1 s, 1/β=2 s, p=.5%, q_i=Ai⁻⁴.

Fig. 8. The density of the stationary throughput (19) conditioned by the fact that the flow is ON. Here R=1 s, 1/μ=100, the loss probability is p=5/100 and η=13. The rightmost mode of the density (around 13) stems from the jump associated with the slow start; the small bump around 7.5 is the “replica” of the rightmost mode (due to the halving of the window).

Fig. 9. The density of the stationary throughput (19) conditioned by the fact that the flow is ON. Here R=1 s, 1/μ=5 and p=5% and η=13.

Fig. 10. The density of the stationary throughput (21). Here R=0.1 s, 1/β=2 s, 1/μ=100 and p=1%.

Fig. 11. The density of the stationary throughput (21). Here R=0.1 s, 1/β=2 s, 1/μ=1000 and p=1%.