An IP-TV P2P streaming system that improves the viewing quality and confines the startup delay of regular audience

Abstract

This paper investigates how a p2p television platform can take advantage of the presence of frequent channel viewers to grant them a more satisfying service than to less regular spectators, and how such privileged users can also be protected when unfavorable operating conditions manifest within the overlay. The explored idea is to learn beforehand about the users’ interests, monitoring their behavior in the recent past, in order to cluster them in groups that display different habits; then, the video chunk scheduling algorithm of the overlay is altered, with the aim of serving frequent spectators in a privileged manner, providing them with a better viewing experience and a faster access to the selected channel without overly penalizing less habitual customers. An analytical model is developed, to capture the difference in average startup delay that the proposed modifications introduce; several additional performance metrics are numerically determined, in order to thoroughly size up the performance of both groups of viewers. The obtained results show that a clear service differentiation is attained and demonstrate that, if bandwidth resources become scarce, frequent viewers are protected and still enjoy a broadcasting service of superior quality.

Appendices

Appendix I

It is first assumed that the system of Fig. 3 is non empty: conditioning upon such event, Fig. 12 allows to conclude that the only observable interdeparture time coincides with the service time of a high interest customer. Hence the conditioned Laplace transform Φ^∗(s) of the p.d.f. of time d is

$$ {\Phi}^{\ast}(s)|\mathrm{non~empty~system}=\left({ \frac{\mu}{\mu+s}}\right) \, . $$

(31)

If the system is empty, Fig. 13 indicates that the interdeparture time is given by the sum of two independent random variables: irrespective of the type of customer who will get into service, the first is a residual interarrival time, exponentially distributed with parameter λ = λ _H + λ _L , the second is a service time, so that

$$ {\Phi}^{\ast}(s)|\mathrm{empty~system}= \left({ \frac{\lambda}{\lambda+s} }\right) \left({ \frac{\mu}{\mu+s}}\right) \, . $$

(32)

Fig. 12

Interdeparture time in the system that differentiates between users, busy server

Fig. 13

Interdeparture time in the system that differentiates between users, idle server

Unconditioning with respect to the server status, which is idle with probability p ₀ and busy with probability 1 − p ₀, Φ^∗(s) expression as appearing in Eq. (13) is attained.

Appendix II

For the system whose state diagram appears in Fig. 3, consider $t_{s_{H}}$, the time spent in system by a high interest customer, and recall that $S^{\ast }_{H}(s)$ indicates the Laplace transform of its p.d.f. Conditioning upon the number k found in system upon the arrival of the high interest customer, when k = 0$t_{s_{H}}$ is the service time of the high interest customer; when k = 1, it is the residual service time of the customer found in service upon the arrival plus one service time; for k ≥ 1, it is one residual service time plus k full service times, each exponentially distributed with parameter μ, so that the conditional Laplace transform of $t_{s_{H}}$ p.d.f. is

$$ S^{\ast}_{H}(s|k) = \left (\frac {\mu} {\mu+s} \right )^{k+1} \, ; $$

(33)

unconditioning with respect to k, $S^{\ast }_{H}(s)$ is determined as

$$ S^{\ast}_{H}(s) = \sum\limits_{k=0}^{\infty} S^{\ast}_{H}(s|k) \cdot p_{k} \, $$

(34)

and replacing p ₀ and p _k , k ≥ 1, as provided by Eqs. 11 and 12 in 34, after some standard, algebraic passages $S^{\ast }_{H}(s)$ turns out to be

$$ S^{\ast}_{H}(s) = \left (\frac {\mu} {\mu+s} \right )c_{1}+ \left (\frac {\mu-\lambda_{H}} {\mu-\lambda_{H}+s} \right )c_{2} \, , $$

(35)

where c ₁ and c ₂ expressions appear in Eq. (27) of Section 4.