Section: New Results

Content-Oriented Systems

Participants : Sara Alouf, Eitan Altman, Konstantin Avrachenkov, Philippe Nain, Giovanni Neglia, Dimitra Tsigkari.

In-network caching is a widely adopted technique to provide an efficient access to data or resources on a world-wide deployed system while ensuring scalability and availability. In previous years, S. Alouf and N. Choungmo Fofack (former PhD student at Maestro , currently at Ingima) have focused on hierarchical systems that rely on expiration-based policies to manage their caches. Each cache in the system maintains for each item a timer that indicates its duration of validity. The Domain Name System (DNS) is a valid application case. The objective was to assess the performance of a polytree of caches. This work has now been published in [4].

In [46], [60], G. Neglia and D. Tsigkari, in collaboration with D. Carra (Univ. of Verona), M. Feng (Akamai Technologies), V. Janardhan (Akamai Technologies) and P. Michiardi (Eurecom), present a new cache replacement policy that takes advantage of a hierarchical caching architecture, and, in particular, of access-time difference between memory and hard disk. They prove that the proposed policy is optimal when requests follow the independent reference model, and significantly reduces the hard-disk load, as they show through their realistic trace-driven evaluation.

Cache networks are one of the building blocks of information centric networks (ICNs). Most of the recent work on cache networks has focused on networks of request driven caches, which are populated based on users requests for content generated by publishers. However, user generated content still poses the most pressing challenges. For such content timelines are the de facto sharing solution. In [53], A. Reiffers-Masson (PhD student in Maestro at the time of submission) and E. Altman in collaboration with E. Hargreaves, W. Caarls and D. Sadoc Menasché from UFRJ (Brazil) establish a connection between timelines and publisher-driven caches. We propose simple models and metrics to analyze publisher-driven caches, allowing for variable-sized objects. Then, we design two efficient algorithms for timeline workload shaping, leveraging admission and price control in order, for instance, to aid service providers to attain prescribed service level agreements.

The non-cooperative nature of relations between economic actors in todays networks may lead to inefficiencies and may not provide incentives for investing in deploying new technologies. In [36] E. Altman in cooperation with V. Douros and S. Elayoubi (Orange Labs) in collaboration with Y. Hayel (UAPV) have studied the question of how to split costs for deploying caches between Content Providers and Internet Service Providers. They have designed the cost sharing by casting the problem into a coalition game which they solved using the Shapely value concept.

The Quality of Experience (QoE) of streaming service is often degraded by frequent play-back interruptions. To mitigate the interruptions, the media player prefetches streaming contents before starting playback, at a cost of initial delay. In [23], Y. Yu and Y. Yu from Fudan Univ. in collaboration with S. Elayoubi (Orange Labs) R. El-Azouzi (UAPV) and E. Altman, study the QoE of streaming from the perspective of flow dynamics. Firstly, a framework is developed for QoE when streaming users join the network randomly and leave after downloading completion. We model the distribution of prefetching delay using partial differential equations (PDEs), and the probability generating function of playout buffer starvations using ordinary differential equations (ODEs) for constant bit-rate (CBR) streaming. Explicit form starvation probabilities and mean start-up delay are obtained. Secondly, we extend our framework to characterize the throughput variation caused by opportunistic scheduling at the base station, and the playback variation of variable bit-rate (VBR) streaming. Our study reveals that the flow dynamics is the fundamental reason of playback starvation. The QoE of streaming service is dominated by the first moments such as the average throughput of opportunistic scheduling and the mean playback rate. While the variances of throughput and playback rate have very limited impact on starvation behavior in practice.