Accessibility setting

Section: New Results

Content-oriented systems

Participants : Konstantin Avrachenkov, Nicaise Choungmo Fofack, Delia Ciullo, Philippe Nain, Giovanni Neglia, Marina Sokol.

Performance analysis of peer-assisted Video-on-Demand (VoD) systems

In [88] and [97] , D. Ciullo, V. Martina and E. Leonardi (Politecnico di Torino, Italy), M. Garetto (Università di Torino, Italy), and G. L. Torrisi (CNR, Italy) consider peer-assisted Video-on-Demand systems. Some of the essential aspects of such systems are peer churn, bandwidth heterogeneity, and Zipf-like video popularity. The authors propose an analytical framework to tightly characterize the scaling laws for the additional bandwidth that servers must supply to guarantee perfect service, taking into account these essential aspects.

The results in [88] and [97] reveal that the catalog size and the content popularity distribution have a huge effect on the system performance. Also, users' cooperation can effectively reduce the servers' burden for a wide range of system parameters, confirming it as an attractive solution to limit the costs incurred by content providers as the system scales to large populations of users. Moreover, in [89] the same authors provide important hints for the design of efficient peer-assisted VoD systems under server capacity constraints.

Analysis of TTL-based cache networks

N. Choungmo Fofack, P. Nain and G. Neglia, together with D. Towsley (Univ. of Massachusetts at Amherst, USA) introduced in [87] a novel Time-To-Live (TTL) replacement policy to manage a set of documents buffering routers in information-centric networks. The TTL policy assigns a timer to each content stored in the cache and redraws the timer at each content request. In [53] they have showed that this TTL policy is more general than other policies like least frequently used (LRU), first-in-first-out (FIFO) or random (RND) as it mimics their behavior under an appropriate choice of its parameters. While exact formulas for the performance metrics of interest (hit/miss processes) are derived for a linear network and a tree network with one root cache and $N$ leaf caches, for more general networks, an approximate solution is found with relative errors smaller than ${10}^{-3}$ and ${10}^{-2}$ for exponentially distributed and constant TTLs respectively. It is demonstrated in [53] that the TTL model can be implemented and used to optimize a multi-content cache network under realistic constraints such as the cache size limitation.

CCN interest routing as multi-armed bandit problem

In [49] K. Avrachenkov and P. Jacko (BCAM, Spain) consider Content Centric Network (CCN) interest forwarding problem as a Multi-Armed Bandit (MAB) problem with delays. The authors investigate the transient behaviour of the $ϵ$-greedy, tuned $ϵ$-greedy and Upper Confidence Bound (UCB) interest forwarding policies. Surprisingly, for all the three policies very short initial exploratory phase is needed. It is demonstrated that the tuned $ϵ$-greedy algorithm is nearly as good as the UCB algorithm, commonly reported as the best currently available algorithm. The uniform logarithmic bound for the tuned $ϵ$-greedy algorithm in the presence of delays is proved. In addition to its immediate application to CCN interest forwarding, the new theoretical results for MAB problem with delays represent significant theoretical advances in machine learning discipline.

In [46] K. Avrachenkov together with L. Cottatellucci and L. Maggi (both from Eurecom, France) consider the choice of CCN Access Points (APs) when CCN APs are wireless base stations. It is assumed that the slow fading channel attenuations follow an autoregressive model. In the single user case, the authors formulate this selection problem as a restless multi-armed bandit problem and propose two strategies to dynamically select a band at each time slot. The objective is to maximize the SNR in the long run. Each of these strategies is close to the optimal strategy in different regimes. In the general case with several users, the authors formulate the problem as a stochastic game with uncountable state space, where the objective is the SINR. Then the authors propose two strategies to approximate the best response policy for one user when the other users' strategy is fixed.