Lieu:amphithéâtre Jade à Télécom ParisTech (46,rue Barrault –75013 Paris)
Membres du jury:
- Christian BONNET(Eurecom)
- Vicente CASARES-GINER (Universidad Politécnica de Valencia,Espagne)
Examinateurs:
- Jean-Marie BONNIN (Télécom Bretagne)
- Geng Sheng KUO (National Chengchi University,Taiwan)
- David BINET (France Télécom)
Directeurs de thèse:
- Daniel KOFMAN (Télécom ParisTech)
Title: Sélection de Réseau dans les Réseaux Sans Fil Hétérogènes
In the context of present trend towards ubiquity of networks and global mobility of services,we notice that network access can be based on a large diversity of access technologies with overlapped coverage. These access technologies might be used simultaneously,which compose an environment of heterogeneous wireless networks. Within this environment,mobile terminals (e.g. smart phones and laptops) and mobile networks (e.g. the network formed by a mobile router in a bus or a train) usually have more than one way to connect to the Internet. If they have protocols to support themselves connecting to the Internet through multiple IP addresses (usually multiple access interfaces),they are called multi-homed mobile nodes (MNs). While if they can only use one IP address through certain access interface at one time,they are called multi-mode MNs. When there are multiple wireless networks,using a wrong network might result in much inconvenience,such as lack of bandwidth for video applications,poor mobility support for high-speed terminals,large signaling cost,traffic congestion,etc. Hence,both multi-homed and multi-mode MNs require to be always connected to the best network and access technology at any time anywhere,which is well known as ‘always best connected (ABC)’.
With ABC functionality,MNs could select appropriate access networks for various applications to fit for their QoS requirements;MNs could avoid selecting a network with high traffic load hence avoid congestion;MNs could predict networks’ availabilities so that they do not connect to networks which disappear soon;MNs could optimize signaling costs by designing their own network selection and handover strategies. In a word,ABC brings plenty of advantages to customers. Moreover,ABC benefits operators too. Due to traffic load assignment control feature of ABC,operators could maximize the usage of their networks hence maximize revenue;based on network selection strategies,operators could analyze and decide their policy on deploying more WiFi access points to attract more customers to WLANs. Finally,ABC is suitable to synthetically consider customers’ and operators’ benefits,so that a win-win partnership can be achieved.
To realize ABC,information about all the available networks and terminal-side entities should be gathered,and the best network should be selected. In the literature,various network selection schemes and mathematical models have been proposed for this task.
In this dissertation,existing network selection schemes,which use various mathematical models,are firstly surveyed. There are also integrated schemes which combine and take advantage of multiple mathematical theories. By analysis and comparison among these models,multi-attribute decision making (MADM) is considered as an appropriate mathematical model for the network selection issue.
Then,a network selection simulator is established by Matlab,which is capable of simulating a large number of scenarios,and it will be used frequently during the further study. Simulation results show that MADM mathematical model works well in plenty of scenarios. Moreover,some important observations are summarized and several existing issues of network selection schemes using MADM are identified. Main issues addressed in this dissertation include:usage of mobility-related factors,requirement of subjective weighting method for evaluating networks,traffic load assignment during network selection,vertical handover (VHO) tradeoff to decide whether a better network is worth handing-over to,etc. First,the issue of using mobility-related factors in the MADM-based network selection framework is studied. If these factors are not used or not correctly used,networks with poor mobility support capability might be selected for a high-speed MN,which disturbs its live applications and increases signaling cost. The proposal in this dissertation can be divided into two steps:a first scenario with two network clusters;a second scenario,more generic,with n network clusters. Since VHO properties are related with permutation of networks,the selection of the best network becomes the selection of the best permutation when VHO properties are taken into account. And,network selection approaches for both scenarios are described and methods to get rapidly the best permutation for the generic scenario are proposed.
Second,the requirement of a new subjective weighting method is analyzed. Network selection is a fast and automatic procedure,which means the traditional analytical hierarchy process method is not suitable for calculating the subjective weights of various attributes in network selection schemes. Hence,a TRigger-based aUtomatic Subjective weighTing (TRUST) method is proposed,which can efficiently calculate subjective weights. In a network selection scheme,the weights of attributes only
need to be evaluated when certain event triggers the network selection procedure,so TRUST calculates weights based on the requirements of the current trigger event. For example,video streaming applications require large bandwidth,so the weight of bandwidth should be large when the network selection procedure is triggered by a video streaming application. Similarly,terminals with high speed require a network with good mobility support feature,so mobility-related attributes should have large weights
when an event ‘terminal speed becomes high’ triggers the network selection procedure.
Third,mobility signaling cost in HMIPv6 networks is evaluated,which shows the way to evaluate total handover cost in the proposed best permutation network selection scheme. Meanwhile,based on the evaluated mobility signaling cost,a mobility anchor point (MAP) selection scheme for HMIPv6 networks is proposed. This scheme minimizes the total additional cost of HMIPv6,including mobility signaling cost and packet
tunneling cost.
Finally,several other issues on network selection are analyzed and possible solutions are proposed,including traffic load assignment during network selection,vertical handover tradeoff schemes,etc. Based on all the studies above,I propose a four-step integrated strategy for MADM-based network selection,which takes advantage of mobility-related factors,uses our efficient weighting method,combines traffic load assignment method,and performs VHO tradeoff before handing-over to the best network.
