Optimal Content Management and Dimensioning in Wireless Networks
Action line : ComEx, task 2|
Subject : Optimal Content Management and Dimensioning in Wireless Networks
Directeurs : Marco DI RENZO, Anastasios GIOVANIDIS
Institution :L2S, LTCI
Phd Student : Jonatan KROLIKOWSKI
Beginning :automne 2015
In recent years the mass of data created and exchanged has vastly increased. Content has become the holy grail of today’s networks, the user’s level of satisfaction depending strongly on the plurality and quality of content received. Consequently, content has become the focus of design in contemporary communications.
Furthermore, wireless networks constitute an important element of future content delivery networks. These are planned to serve a constantly increasing number of internet users asking for connectivity and access everytime/everywhere. The current design trend for wireless networks is densification, based on the idea that higher throughput can be achieved by deploying additional nodes.
In such dense topologies, the amount of traffic will soon be too much to be served by single source nodes. On the contrary, performance improvements can be achieved when many copies of the same content are spread throughout the network, increasing the possibility to find one close to the requesting user. This idea has already started being studied and has led to the Information Centric Network (ICN) proposal.
A natural extension - not well studied until today - is to exploit dense wireless environments and add large memory to their nodes where popular content can be cached. In this way, part of the demand will be served from close-by nodes without overburdening the backhaul. The user experience will thus improve, since service will suffer less delay and buffer starvation issues.
Caching in general has seen a revival of interest especially due to the concept of the ICN architecture (see e.g. the survey in 2). However, its extensions to wireless networks have not yet been enough investigated. It is this gap that the suggested thesis wishes to cover. Why is this interesting ? Because wireless networks have specificities that render existing results for caching in wired not sufficient :
The aim of the thesis is to propose and evaluate the performance of realistic and (near-)optimal cache management policies and dimensioning rules (memory size and placement), that can improve both the backhaul traffic and the Quality-of-Service/Quality-of- Experience (QoS/QoE) of the served wireless user.
Specifically, the involved doctorate student will :
- 1. Propose, analyse and evaluate novel caching policies for wireless networks that can improve performance over the existing ones (compared to Least Recently Used (LRU), Probabilistic Least Recently Used (q-LRU), Least Frequently Used (LFU), full information policies). Use different performance measures additionally to the hit ratio.
- 2. Dimension appropriately the cache equipped network. This part will consider the optimal choice of cache-memory size as well as the necessary number and choice of positions to install caches to for specific QoS/QoE guarantees. A step further could consider the possibility to install additional (how many/where) cache-equipped small BSs to improve coverage.
- 3. Understand the traffic characteristics of wireless networks and build caching policies based on stationarity and non-stationarity assumptions of the traffic. Make use of learning-based approaches to adapt the content estimation to the traffic evolution.
- 4. Find the trade-offs when a user has to choose between content availability in cache or better QoS provided (higher SINR, lower delay, jitter) from his/her connected base station.
- 5. Further investigate the influence of caching on issues of QoE (Quality-of-Experience), like multimedia loading time, buffer starvation events, etc. Within this, take the issue of user mobility into consideration and how the user trajectory should influence content placement strategies.
For traffic and topology modelling, tools from stochastic analysis will be used (random processes in 1D and 2D). The student will propose and analyse future strategies using techniques from optimisation (convex, discrete) and potentially, algorithms from the machine learning literature, in order to deal with unknown traffic demand distributions.