Novel Mobility Management Algorithms

Transcription

1 Budapest University of Technology and Economics Department of Telecommunications Novel Mobility Management Algorithms PhD. Thesis Szalay Máté Research supervisor: Dr. Imre Sándor Budapest, 2006.

2

3 I hereby certify that this material, which I now submit for assessment on the programme of study leading to the award of PhD is entirely my own work and has not been taken from the work of others save and to the extent that such work has been cited and acknowledged within the text of my work. i Alulírott Szalay Máté kijelentem, hogy ezt a doktori értekezést magam készítettem, és abban csak a megadott forrásokat használtam fel. Minden olyan részt, amelyet szó szerint, vagy azonos tartalomban, de átfogalmazva más forrásból átvettem, egyértelműen, a forrás megadásával megjelöltem. Budapest, Szalay Máté A dolgozat bírálata és a védésről készült jegyzőkönyv a későbbiekben a dékáni hivatalban elérhető.

4 ii

5 Contents 1 Introduction Motivation Background Terminology An Engineering Problem Research Goals Universal Location Management Model Novel Mobility Management Mechanisms Research Methodology Structure Related Work IP Mobility Mobile IPv Mobile IPv Markov Chains Location Management Hierarchical Location Management Paging Location Management Algorithms Cellular Network Structure Models Modelling the MN movement and packet delivery Mobility Management Classification Introduction Mobility Management Hierarchy Single-layer (Flat) Solutions Exact Location is Known Exact Location is not Known Summary of Single-layer Solutions Hierarchical Solutions Two Layers More Layers iii

6 CONTENTS iv Summary of Multi-layer Solutions Empty Classes LTRACK Introduction LTRACK Network Mechanisms Locating the Mobile Node Handover Qualitative Analysis Simulation LTRACK vs. Other Protocols Limiting the Number of Tracking Handovers Using LTAs Limiting the Number of Visited Nodes Analytical Model LTRACK Network Model Parameters LTRACK Over Tree Topology Network Cost Function Handover Model Modelling the existing solutions Optimizing LTRACK Analysis and Comparison Summary Hierarchical Paging Introducing Hierarchical Paging The Approach Mechanisms Qualitative Considerations Analytical Model Network Model Mobility Model Markov-model Changing the Parameters The Effect of the Number of Subnetworks (k) Flat vs. Hierarchical Paging The Effect of the Size of the Network (L) The Effect of the Mobility Ratio (p) Simple Simulations Simulation Environment Simulation Goals

7 CONTENTS v Simulation Parameters Simulation Results Advanced simulations Simulation Environment Simulation Goals Simulation Parameters Simulated Algorithms Comparative Simulation Cost Coefficients Special Considerations More Layers A Specific Problem Dynamic Allocation of Layers Summary Conclusions and Future Work Conclusions Application of the Results Future Work List of Figures 80 List of Abbreviations 82 References 83 Publications 87

8

9 Chapter 1 Introduction 1.1 Motivation Over the past few years there has been extreme growth in wireless communications. While mobile telephony remains one of the most widespread mobile application, mobile web browsing, sending and receiving or even mobile banking are becoming more and more widely used. There is a fierce competition among service providers, and efficiency is very important factor in this competition. A provider that can use its infrastructure in a more efficient way has a competitive advantage over the others. 1.2 Background Any mobility protocol has to solve two separate problems: location management (sometimes called reachability) and session continuity (sometimes referred to as handover management). Location management means keeping track of the positions of the mobile nodes in the mobile network, session continuity means to make it possible for the mobile node to continue its sessions (e.g. phone calls) when the mobile node moves to another cell and changes its service access point. In my thesis I am concentrating on location management and not session continuity. It is important to note, that for this reason the terms location management and mobility management are used interchangeably. Location management has to address the following questions[4]: When should the mobile terminal update its location to the network? 1

10 CHAPTER 1. INTRODUCTION 2 When a call arrives, how should the exact location of the called mobile equipment be determined? How should user location information be stored and disseminated throughout the network? These problems are usually solved in two stages: location registration (or update or tracking) and call delivery (or searching)[1]. The solution usually use a hierarchical approach, see Section Because of the growth of mobile communications and the limitations of resources (especially frequency), more and more efficient algorithms are needed for routing, call management and location management. 1.3 Terminology In a wireless network service access points are usually called base stations. The node that is moving around in the mobile network is called mobile node or mobile equipment. Mobile nodes are connected to the network via base stations, each of the base stations covers one cell, and there are wireless links between the base stations and the mobile equipments. Base stations are interconnected with routers to form a network; this network usually uses fixed and sometimes microwave links. As the mobile node is moving around, it changes its point of connection to the network from time to time. The event, when the mobile equipment moves to a new service access point is called handover or handoff. In a mobile network the terms location and position usually do not mean geographical position, but they refer the location of the mobile node within the network, the service access point (base station) it is connected to at a specific moment of time. Although there are exceptions (like Location Based Services, LBS where the geographical location is important), throughout this paper both position and location refer to the location from the network point of view. To let the network have location information about a specific mobile node, location update messages are sent to the network. Location updates are usually sent at handovers but some protocols may require them to be sent periodically even if no handover takes place. The network doesn t always have exact location information about every mobile. If a

11 CHAPTER 1. INTRODUCTION 3 mobile has to be found in the network, and its exact location is not known, searching takes place. See Chapter 3 for a detailed analysis of update and searching. 1.4 An Engineering Problem In a mobile network both tracking and searching have a cost. Here, cost doesn t necessarily mean a cost paid in money, it refers to a general price that is to be payed in money, time, algorithm complexity, wasting of resources, shorter equipment battery lifetime, lower user satisfaction, etc. Consider one scalar parameter of a mobility management solution. If an increase in the parameter value increases some cost factors and does not change others on the whole range, then there is no conflict: the lower parameter value, the better. If changing the parameter might increase some cost factors and decrease others then there is a conflict which might lead to an optimization problem. In this sense, there is a general conflict between tracking and searching. The scalar parameter is the frequency (and perhaps accuracy) of location updates. If more update messages are sent then the network will have more accurate position information of the mobile node most of the time. On one hand, update messages have a cost, but on the other hand, if we have more accurate position information, the (expected) cost of the searching procedure will be smaller. The price of sending updates more frequently is paid in higher signalling load on the network, higher communication channel usage, shorter battery life for the mobile node. The price of searching is paid in more signalling, more communication channel usage, longer search procedures (lower user satisfaction), and perhaps higher algorithm complexity. Thus, updating more frequently makes the total cost of updates higher, but the total cost of searching lower; and updating less frequently makes the total cost of updates lower, but the total cost of searching higher. Of course this is just an oversimplified model of the problem of location management, but it is easy to see that depending on how various costs (time, money, signalling load) are compared to each other, this can lead to optimization problems. There can be optimum update frequencies which lead to the lowest total (update and searching together) cost possible. My thesis addresses this problem.

12 CHAPTER 1. INTRODUCTION Research Goals Universal Location Management Model There are several location management algorithms in the literature. For a good survey on them, see [1]. The first research goal was to provide a universal model, or classification system for location management. The model should be general enough that it covers most (if not all) of today s location management algorithm, and should enable us to classify and compare various algorithms Novel Mobility Management Mechanisms To make mobility management more efficient, we can either tune the parameters of an existing algorithm to suit the environment better or completely new algorithms can be designed. The most important research goal was to find novel mobility management mechanisms and build algorithms that use them. The efficiency of an algorithm depends on several factors, so we usually can t say that an algorithm is better than another one in every scenario and at every parameter setting, this was not my goal. My goal was to find new location management algorithms that outperform most of today s solutions at reasonable parameter settings of reasonable scenarios. 1.6 Research Methodology To find out if my new location management methods are indeed more efficient than others, I had three approaches: analytical considerations, simulations and measurements. When developing novel algorithms, in the early phase usually simulations and analytical methods are used; those are the approaches that I have used in my work. To use the third approach measurements my algorithms would have had to be implemented, which is usually a time (and money) consuming process, which should normally follow the previous two. To use an analytical method, first a mathematical model had to be constructed. I have defined the mathematical model for both of my new algorithms. Both of the models make use of Markov-chains extensively. For the mobile network I used a simple graph model, the mobility model was taken from literature with some minor extensions. I have also created several simulations. I haven t used any specific simulation tools, but built my own simulation framework in C++. Some of the simulations are created using

13 CHAPTER 1. INTRODUCTION 5 MATLAB[17], others are built upon my own simulation framework. 1.7 Structure This dissertation is structured as follows: After the introduction, in Chapter 2 the related work is presented very briefly. Mobility and location management are explained in general, and some mathematical modelling techniques, such as Markov chains are introduced. A brief survey is given on state-of-theart location management methods which are related to or can be compared to my own algorithms. In Chapter 3 I introduce my own classification system of mobility management algorithms. This is a general classification method that can be used to classify existing and new algorithms and which has led to the invention of Hierarchical Paging. LTRACK, my first mobility management algorithm is introduced in Chapter 4. First the network architecture and mechanisms are described, then qualitative analysis, simulations and quantitative analysis are presented. My algorithm is compared to other location management algorithms, and it is examined how changes in various parameters affect the performance of LTRACK. My second mobility management algorithm called Hierarchical Paging is presented in Chapter 5. After qualitative considerations an analytical model is presented for it, which is compared to my simulations of Hierarchical Paging. The analytical results and simulation results fit very well. Simulations are also used to compare Hierarchical Paging to other location management algorithms, and the parameter ranges where Hierarchical Paging is most efficient. Finally, in Chapter 6 the conclusions are drawn, possible applications of my result and possible future directions in this research area are presented.

14 Chapter 2 Related Work 2.1 IP Mobility This section gives a short introduction to IP mobility, possible solutions and potential problems. The purpose of the section is to show how the general principles and mechanisms introduced in Chapter 1 correspond to the principles and mechanisms of a real-world protocol Mobile IPv4 Regarding the static Internet, IP packets are routed depending only on their IP destination address. In this manner, the IP address is used to route the packet. On the other hand, it is also used to identify a host or network. Having a mobile environment, the node identified by an IP address might move away while the packets are still delivered along the same static route. To resolve this problem in the Mobile IP protocol, two IP addresses are used: one to identify the host and another one for routing purposes. With this strategy the identification role of the IP address is kept while the other address can be used for routing. The protocol consists of three different sub-protocols. The Agent Discovery is where Mobility Agents advertise their availability on each link on which they provide service. After receiving such an advertisement the MN decides whether it is in the Home Network or in a Foreign Network. (We do not discuss the case when the HN is detected since then no mobility support is needed.) For the second step, the Mobile Node Registers its Care-of Address with its Home Agent. This is an update or tracking procedure The CoA is obtained from the FN either with any external assignment 6

15 CHAPTER 2. RELATED WORK 7 mechanism or with the advertisements received. For more details on Care of Addresses, see [22]. When a correspondent node wants to communicate with the MN, it sends the datagrams to its Home Address. The Home Agent intercepts the communication and tunnels the datagrams to the Care-of Address of the MN. The endpoint of the tunnel can be either a Foreign Agent or the Mobile Node itself. If it is delivered to a FA, the FA has to forward the packets to the MN. As the location of the mobile node is always exactly known to the network (Home Agent and Foreign Agent), no searching is required. When the MN replies to the Correspondent Node it sends the packets directly to it. This is called the Triangle Routing. (It is useful to note that if the CN is a mobile node as well, the protocol automatically works since having a mobile or a static node as a peer is indifferent. This can be considered as one of the main advantages of MIP.) This routing system is far from optimal and there are many proposals to make it more effective. (One can think about an MN near to the CN but roamed to a network far form its HN.) Updating Mobility Bindings by providing the CN with the CoA of the MN would shorten the path for the datagrams to be delivered. This is called route optimization and unfortunately has serious security issues Note that route optimization can be considered as a special update mechanism. It is also clear that the CN has to be capable to this protocol that is evidently a disadvantage of such an extension. Also there is the problem of smooth handoffs that has to be solved and additional bindings from the FA to the MN has to be applied. This is the domain of session continuity. Tunnelling the packets from the HA to the CoA also provides basic security for the communicating entities since the real destination address (the Home Address of the MN) is hidden from any intervening routers. This is useful because the location of the MN is hidden from many unauthorized networks elements. However, security issues still need further considerations since a corrupt FA can still easily access to possibly confident information Mobile IPv6 The Mobile IPv6 protocol [10, 6] provides IP mobility over IPv6. It uses the advantages of the newer IPv6 protocol and also the experiences gained from the Mobile IPv4 are taken into consideration. As it is described earlier, in case of Mobile IPv4, when the Mobile Node is away from home, a Care-of Address is assigned to it, and can be reached transparently on his Home Address.

16 CHAPTER 2. RELATED WORK 8 When the Mobile Node gets new CoA, it registers a Binding with a router in his home network. This association is recorded in the Binding Cache of the router. The MN now can be reached through the HA with IPv6 encapsulated, tunneled packets. The implementation of this mechanisms is part of reachability. The MN can also send Binding Update to the Correspondent node and with this additional binding the packets sent can be delivered directly to the MN. During this operation authorization of the MN and the CN is done using the IPv6 header, this mechanism can be considered a special case of reachability of location management. A loss of these binding updates may occur when the CN is moving as well. If we take a short look at such a scenario, it can be seen that the Binding Update sent to a moving node might arrive to the former location of it and the communication between the two nodes will be lost [30]. 2.2 Markov Chains Using Markov chains is a widely used technique for mathematical representation of problems in different fields of telecommunications[13]. For an introduction to Markov chains see Chapter 11 of [5]. I used discrete-time and continuous-time, finite-state and infinite-state Markov chains to model the handover processes and various mechanisms of LTRACK (Section 4.6.5) and Hierarchical Paging (Section 5.3.3). 2.3 Location Management Hierarchical Location Management Even if the covered area, where mobility is provided is not too large, mobility management is often handled in a hierarchical way for efficiency and scalability reasons[1]. As the characteristics that are observed at various levels can be quite different, it is not surprising, that the different solutions handle mobility the most efficiently at the different levels. This means that there is a hierarchy of different mechanisms working together the provide the most efficient solution for a given environment. For a detailed analysis of this area see Section

17 CHAPTER 2. RELATED WORK Paging If the exact position of the mobile is not known to the network, and it has to be determined, searching takes place. A widely used search mechanism in cellular networks is called paging. It means that paging messages are broadcasted to the mobile nodes, and the mobile that the the network is paging has to reply the paging message. There are various strategies for paging, for references see Section 2.3.3, for a more detailed explanation of paging see Section Location Management Algorithms Neither the centralized approach of Mobile IP nor a naive paging approach is efficient enough on its own when covering a large area and providing services for several mobile users. There is an increasing number of solutions for both location management and handover management in mobile networks that address this problem. Most of them use a hierarchical approach, and some introduce special mechanisms at various levels. For some IP (Internet Protocol) based solutions see [4, 12, 26, 28]. Good general surveys on the topic are [31, 34]. These solution usually use a hierarchical approach, but none of them use location tracking, the main idea behind LTRACK, see Chapter 4. Several studies have addressed paging, various paging methods are presented, usually considering cost minimizations. Almost all of these papers focus on the paging algorithm itself, but the paging is always just at a single (usually the bottom) level of the network (flat paging, see Chapter 3). Ramjee et al. examine and compare three different paging architectures and protocols [15], all flat. Hajek, Mitzel and Yang [2] show algorithms to optimize registration and paging together. They examine serial and parallel paging, but none of their solutions introduces a hierarchy in the paging itself as my solution (Hierarchical Paging, see Chapter 5) does. Usually, paging is used at the bottom level of a hierarchical mobility management solution, as in case of a GSM network [7, 25]. In their paper[29], Woo et al. optimize location management for a special types of networks, namely Two-Way Messaging networks. The paper of Akyldiz et al.[1] is a good general survey on various existing and proposed mobility management algorithms. None of the above solutions use or mention Location Tracking or Hierarchical Paging

18 CHAPTER 2. RELATED WORK 10 as an alternative solution. An approach similar to LTRACK is used by RoamIP [28] but only for session continuity and not location management purposes. 2.4 Cellular Network Structure Models Hexagonal cell structure is the most widely used model for cellular coverage. Figure 2.1 shows a cellular network with hexagonal cells and a subnetworks. The shape of the cells is close to the natural circle-shape, and each cell has six neighboring cells. Figure 2.1: Subnetwork with hexagonal cells For some simulations, a simplified cellular network structure was used where the area is covered using square-shaped cells, as in Figure 2.2. In this case, each cell has four neighboring cells, this model is less realistic than the hexagonal one, but it is more easy to handle it mathematically.

19 CHAPTER 2. RELATED WORK 11 Figure 2.2: Subnetwork with square shaped cells In both cases the cells can be grouped to form subnetworks. Subnetworks are usually logical and not geographical groups of cells, so cells belonging to the same subnetwork do not have to be neighboring (or even geographically close to) each other. Cells are often grouped to subnetworks to form paging areas (or for other hierarchical mobility management purposes), so cells belonging to the same subnetwork almost always form a geographically continuous area. 2.5 Modelling the MN movement and packet delivery In this Section, the model for the MN movement and the packet delivery is be presented. It is assumed that in a given time interval the MN changes its point of attachment with a constant rate. This will be modelled with a homogenous Poisson process as in other

20 CHAPTER 2. RELATED WORK 12 works [8, 16]. Let λ denote the parameter of the Poisson process and so denote the rate of handovers of an MN. The other parameter that can be introduced in a similar manner is the rate of receiving a call: µ. This can also be time- or location-dependent but we assume it is constant for the examined time interval. Let us introduce ρ as the mobility ratio meaning the probability that the MN changes its FA before a call arrives. Here ρ = λ. In [8] rho is defined like ρ = 1 α(0). (Note λ+µ that the ρ in the two works are not the same! The ρ in [8] would be ρ = µ with our λ notation.) In real systems it might be not true that the mobility ratio is homogenous. Parameters µ and λ might depend on the position, the specific user, the time of the day, etc. Our model does not handle this scenario. However, we can say that these constant values are some averages, and valid for specific users, at a specific network region in a specific time interval.

21 Chapter 3 Mobility Management Classification 3.1 Introduction In this Chapter my general model of mobility networks is presented. It is a model in the sense that it is independent of technology and actual mobility solution, and it is general in the sense that it covers most of the current technologies and solutions. By using this general model, the different mobility solutions can be classified. After classifying the mobility solutions of today, we may find that some classes are empty. A class may be empty for two reasons: It is impossible or senseless to implement a mobility algorithm that belongs to that class. For example an algorithm belonging to another (specific) class obviously outperforms it in every sense. No mobility algorithm has been invented yet that belongs to that class. The latter case may be exciting; this is the case where our general and solutionindependent mobility model has actually enabled us to invent something new that might be rather specific Mobility Management Hierarchy The area where mobility is provided can vary widely from network to network. A satellitebased system may cover a whole continent (or even a whole planet), while a small-scale, targeted mobile network may cover only a building or a part of a building. 13

22 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION 14 Even if the covered area is not too large, mobility management is often handled in a hierarchical fashion for efficiency and scalability reasons. The area that is covered is divided into smaller areas, the whole network is divided into sub-networks. These subnetworks together cover the whole area, and the areas covered by the subnetworks may even overlap sometimes. The point here is that all mobility within a subnetwork can be handled locally (within the boundaries of that subnetwork) in a completely transparent way for the other parts of the network. The subnetwork then can be again divided into sub-subnetworks. A solution that is not hierarchical is called flat, or single-layer. A hierarchical solution can be more efficient than a flat one by using less resources for mobility management. And at the same time, hierarchical solutions usually scale much better with the size of the network and the number of mobile nodes than flat ones. 3.2 Single-layer (Flat) Solutions In this Section I am going to address flat mobility management solutions. These solutions can either be used as a mobility management algorithm for a single-layer (flat) mobility network or as an algorithm on a specific single level of a hierarchical, multi-level mobility network. The later case is going to be discussed in Section 3.3. So, we are not dealing with hierarchical solutions here. What kind of flat solutions are there? Our classification property will be wether the exact location is always known to the network or not. There are solutions where the network has exact location information of all the mobile nodes all the time, and there are solutions where the exact location is determined in an on-demand way. These solutions are discussed in detail in the next subsections Exact Location is Known This is the more straightforward situation. The exact location of all the mobile nodes is known to the network all the time. To enable the network to know have exact location information, the mobile node has to inform the network when it moves from one access point to another one, so a handover can take place. The messages that are sent between network nodes containing information about the current location of a mobile node are called location update messages. The exact position of a mobile node can be stored in a central database or in a distributed one.

23 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION 15 Centralized Solutions In case of a centralized solution there is an entity (network node) for every mobile node that has exact position information of it all the time. This network node can be the same for all the mobile nodes or there can be several of them scattered across the network; each one serving just a subset of the mobile nodes. An example of the first case is the top-level mobility management of GSM [7, 25], where a network element called Home Location Register (HLR) has up-to-date position information for all the subscribers of a provider. (In fact, GSM has a hierarchical mobility management solution, and this is just the highest level of the hierarchy. The HLR does not know which base station the mobile node is connected to, just which network it is roaming in.) An example of the second case (several position-storing nodes each covering a subset of the nodes) is the MobileIPv4 mobility management scheme[19, 23]. In this solution each mobile node has a so-called Home Agent (HA), which has exact location-information of the mobile node all the time. One HA can serve several mobile nodes, but there can of course be more Home Agents. Actually, there has to be several, because in the MobileIP scheme a HA can only serve mobile nodes that it shares the network part of the IP address with. So mobile nodes with home IP addresses on different networks have to have separate Home Agents. While there are several Home Agents, this is not a distributed solution, because for each mobile node there is one Home Agent which has exact location information all the time. It may be called Home Agent, Home Location Register or something else, the important point here is that there is one central network node for each mobile node which always has to know the exact position of the mobile node. Thus, every time the mobile node moves out from the coverage of a base station, a handover has to take place, and the central network node has to be notified. This solution obviously raises some efficiency and scalability problems. If the central node is very far away (several hops away) from the mobile node, when the mobile node is moving, the location update messages have to travel a long distance, thus they consume a lot of resources and take a lot of time, causing long delays. Another problem is that if a lot of mobile nodes that are served by the same central node are moving intensively, the central node may become overloaded by location update messages.

24 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION 16 Distributed Solutions There are solutions where the exact location of the mobile nodes is known to the network, but there is no single network element which has exact location information for even one of the mobile nodes all the time. We will call these solutions decentralized or distributed solutions. An example of this distributed database case is the LTRACK mobility management method [C9, C11]. LTRACK (Location Tracking) is a mobility management approach introduced in Chapter 4 of this thesis explains LTRACK in detail, so I ll just give a short summary here. An LTRACK network is built up from LTRACK nodes,and the mobile node is connected to one of the LTRACK nodes in the network, and it can change its point of connection. An LTRACK node is a logical entity, it can be a base station of the network, or it can be a small subnetwork where mobility is handled locally (Hierarchical solution with LTRACK at the high level). Every mobile node has an entry in a home LTRACK register (HLR). This register is similar to the Home Location Register of GSM networks or the Home Agent of the MobileIP scheme. The HLR of LTRACK does not have exact location information, but the LTRACK network as a whole has, so when an incoming packet arrives, the exact location of the mobile node can be determined. For each of the mobile nodes, the HLR stores the last address where it received location update message from. It is a next-hop towards the mobile node. The mobile node is either connected to that LTRACK node, or that LTRACK node knows another nexthop LTRACK node towards the mobile. Once an incoming call arrives, there is a series of LTRACK nodes pointing from the HLR to the mobile node, see Figure Exact Location is not Known It is possible to define a single-layer scheme where the location of the mobile node is not always known to the network. The advantage of this scheme is that the mobile node does not have to notify the network every time a handover takes place. Fewer location update messages imply smaller signalling load on the network, and longer battery life for the mobile node. On the other hand, if the exact location is not known, then the mobile node has to be found in some way, in case it has to be contacted (for example an incoming call or packet

25 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION 17 arrives). In cellular networks, the procedure when the mobility network determines the exact position of the mobile node is called paging. Paging messages are broadcasted to the mobile nodes, and the mobile that the the network is paging has to reply the paging message. The simplest solution for paging is to flood the whole network with paging requests (all the base stations broadcast it in all the cells of the network), and have the mobile node answer it with a location update message. Of course this method of paging is very expensive in the sense of signalling load. There are some other, more sophisticated strategies for paging (see Section 2.3.3), but the process is usually very costly (in the sense of signalling load and radio frequency usage) compared to other procedures of mobility management. For example an advanced paging method can first try to find the mobile node in the cell it was last known to stay in, and then move further and further away from that cell until the mobile node is found. If the mobile nod has not moved a long distance since the last location update, it is going to be found without much effort, if it has moved a lot, the procedure might take longer, and consume more resources. The advantage of this method compared to the naive, one-step paging procedure is that it may require much fewer paging messages, the disadvantage is that it may take much longer. A real-life example where paging is used is the lower level of the GSM network architecture [7, 25]. The cells of the network are grouped into location areas (LA). Each cell belongs to exactly one location area, and every base station periodically broadcasts the location area identifier (LAID) of the location area it belongs to. To save the battery and to reduce signalling, the mobile node can switch to idle mode. When in idle mode, the mobile node has to send location updates only when it moves to a new location area, but does not have to notify the network when it keeps moving around within a location area Summary of Single-layer Solutions Table 3.1 shows the attributes of possible single-layer mobility solutions along with some examples. Table 3.1: Single-layer mobility management algorithms Single-layer mobility management Exact location is known Exact location is not known Centralized Distributed MobileIP, Home Agent LTRACK GSM paging

26 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION Hierarchical Solutions Mobility networks that cover larger areas usually don t have a flat structure, but implement some kind of hierarchy. As it was explained in Section 3.1.1, the network is divided into subnetworks. Mobility within a subnetwork (intra-subnetwork mobility) is handled by the lower level mobility management, while inter-subnetwork mobility is handled by a higher level mobility management solution. In this Section I am going to discuss how my general mobility model can be extended to cover these hierarchical solutions too. In my model, top-level mobility management, which handles inter-subnetwork mobility management is considered as a single-layer mobility management algorithm. The only difference is that the mobile node does not move form base station to base station but from subnetwork to subnetwork. The inner structure of the subnetwork and all the events happening and data stored within the subnetwork are hidden from this higher-level mobility management algorithm. Locating the mobile node in a single-layer network means finding which base station it is connected to, locating it in the top-level of the hierarchy means to find which subnetwork it is staying in Two Layers Consider a two-layer mobility network: there is a top layer (or top level) and a bottom layer (or bottom level). Both of the layers are implemented using a single-layer mobility management solution. The specific mobility management solution can be the same for the two layers or it can be different. In this subsection some examples are discussed for this scenario. Centralized Top Level, Paging at the Bottom Level In this example the centralized solution is implemented at the top level, and the position is not always known at the bottom level. It means that the network is divided into subnetworks, and for every mobile node there is a central entity that always knows which subnetwork the mobile node is currently staying in, but within the subnetwork, the exact position is not always known, thus paging is used. This special case is used in real-life, public cellular networks, such as GSM implement this scheme. In GSM terminology, the subnetworks are called location areas (LA).

27 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION 19 Centralized Top and Bottom levels Our model allows the use of the same solution at different levels. If both the top and bottom levels are centralized, there is an entity that always has the information which subnetwork the mobile node is staying at, and for each subnetwork there is an entity that has exact position information for the mobile node as long as it is staying at that subnetwork. Several hierarchical extensions of MobileIP, for example [24] or [9] use this scheme. Paging at Top and Bottom Levels At first it seems that if the location is known neither at the top level nor at the bottom level, and paging is used at both levels than this scheme is no different from the singlelayer paging scheme (see Section 3.2.2). Actually, this is not the case, because although the exact location is not known at the bottom level, it is known wether the mobile is actually within that subnetwork or not. And this makes an important difference. Consider a network where mobility management is handled in this fashion. The network is divided into subnetworks, and there is an entity in each subnetwork that is responsible for the mobility management. This entity is going to be referred to as root node. It either knows the exact location of the mobile node within the subnetwork, or it can determine it by using a paging algorithm. Thus, while the mobile node is moving within a subnetwork, no location updates are sent. There is no handover at all in the top level, so when the mobile node has to be found the top level knows which subnetwork the mobile is currently in, and within that subnetwork paging is used. In this case the gain of the scheme is high over the single-layer paging scheme, because the broadcast of paging messages is limited to the subnetwork. When the mobile node moves to a new paging area, the root nodes of both the old and the new subnetworks have to be notified, but inter-subnetwork mobility can still be handled locally. So the old root node learns that the mobile node is not in its subnetwork anymore, and the new root node learns that the mobile node is in its subnetwork. When an incoming call or packet arrives destined to the mobile node, the old root node is contacted first, because top level mobility management still thinks that the mobile node is in the old subnetwork. But the old root node knows that the mobile node is not there, so a top level paging takes place. This top level paging means that gateway of the network (or other central entity) pages all the root nodes of the subnetworks, and the root node of the new subnetwork is going to give a positive reply. This top level paging is

28 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION 20 still requires much less signalling than a single layer paging, so it is much more efficient. Although this scheme is much more efficient than single-level paging, its scalability might be a problem. If some smart paging algorithm is used at the top-level, this solution might become feasible. To decide wether it is profitable to use hierarchical paging instead of single layer paging various parameters such as mobility intensity, incoming call intensity, various signalling and processing costs have to be taken into account. I know of no implementation of this scheme, this is an example where the general model helped to discover something specific. This scheme, called Hierarchical Paging is described in detail in Chapter More Layers Of course our model for mobility networks can handle more than two layers. If there are several mobility management levels in a network, each level may use any kind of single-layer solution. For example if there are several layers (levels), but a centralized solution is used at all of them, the scheme is very similar to Dynamics Mobile IP [9]. Dynamics Mobile IP introduces a tree topology hierarchy of Foreign Agents between the Home Agent and the mobile node. All the handovers are handled as locally as possible; within the smallest subtree that contains both the old and the new base stations Summary of Multi-layer Solutions Table 3.2 shows the summary of two-layer mobility solutions. Different columns represent different top-level mobility management solutions, different rows correspond to different bottom-level solutions. 3.4 Empty Classes In the classification process, classes were defined, and then mobility management algorithms were put in these classes. Recall from 3.1, that if a class turns out to be empty (ie. there is no algorithm in it), it can mean two things: either it is senseless or inefficient to implement a mobility management algorithm that belongs to that class (for example algorithms in another class clearly outperform the class in every aspect),

29 CHAPTER 3. MOBILITY MANAGEMENT CLASSIFICATION 21 Bottom level Table 3.2: Two-layer mobility management algorithms Top level Centralized (Exact Decentralized Exact location is location is (Exact location not known known) is known) Centralized TELEMIP, Dynamics (Exact location MobileIP is known) Decentralized LTRACK LTRACK (Exact location is known) Exact location GSM LTRACK with Hierarchical is not paging Paging known or no algorithms have been invented so far that belong to that class. My LTRACK mobility management method was not invented this way (it was introduced a year before the classification system), but Hierarchical Paging (introduced in Chapter 5 was actually invented by finding and empty class and then realizing (and proving) that efficient algorithms can be designed that belong to that class.

30 Chapter 4 LTRACK 4.1 Introduction LTRACK (Location Tracking) is my completely new approach for location management in mobile networks. It was introduced in [C9]. The basic idea behind LTRACK is to find a compromise between the Mobile IP scheme (where the HA has exact location information) and the GSM scheme (where only the LA and no further information is known). The exact location of the mobile node is not stored in a central database, but the information is shared among the network nodes that the mobile node had visited recently. The searching runs through the chain of these nodes. [J1, J3, C9] 4.2 LTRACK Network An LTRACK network is built up from LTRACK nodes. A mobile node is connected to one of the LTRACK nodes in the network, and it can change its point of connection. Every mobile node has an entry in a home LTRACK register (HLR). The HLR of LTRACK does not have exact location information, but when an incoming packet arrives, the exact location of the mobile node can be determined. LTRACK nodes can be routers in the real world. How are base stations connected the network of LTRACK nodes? There are three different approaches: Base stations are LTRACK nodes too. Base stations are connected to LTRACK nodes. 22

31 CHAPTER 4. LTRACK 23 Hierarchical approach. The naive solution is to define base stations as LTRACK nodes too. When the mobile equipment moves from one base station to another, it moves from one LTRACK node to another. In this scheme the routing functionality and the base station functionality get mixed up, which is usually undesired. An approach that uses a more structured network is to connect base stations to LTRACK nodes. In this scheme LTRACK nodes are similar to GSM Base Station Controllers (BSC). An LTRACK node can serve several base stations. When a mobile equipment moves from a base station to another one, and both base stations are served by the same LTRACK node, the old and new LTRACK nodes are the same. That is the only LTRACK node that has to be notified. When the old and new base stations are served by different LTRACK nodes, an LTRACK handover takes place. This solution decreases the number of required LTRACK handovers for the same number of handovers. Thus, a smaller number of hops will be needed when trying to find the mobile node. The hierarchical approach is to define LTRACK nodes as small networks. The networks at the lower hierarchy level can be any kind of mobility networks, they can even be LTRACK networks. Thus a two or more level LTRACK network can be built. 4.3 Mechanisms Locating the Mobile Node In LTRACK, each mobile node has a unique identifier similar to IP addresses or phone numbers. This unique identifier is connected to its home address. It is similar to the home address of the Mobile IP scheme. For each of the mobile nodes, the HLR stores the last address where it received location update message from. It is a next-hop towards the node. The mobile node is either connected to that LTRACK node, or that LTRACK node knows a next-hop LTRACK node towards the mobile. Once an incoming call arrives, there is a series of LTRACK nodes pointing from the HLR to the mobile node, see Figure 4.1. LTRACK nodes has to be able to find routes to each other. This can be solved for example by using LTRACK over an IP network, thus letting IP routing do the job.

32 CHAPTER 4. LTRACK Tracking Handover Tracking Handover Figure 4.1: LTRACK locating the mobile node Handover When the mobile node moves from one LTRACK node to another, a handover takes place. The LTRACK node that the mobile moves away from is called the old LTRACK node, the one it moves to is called the new LTRACK node. There are two different kinds of handover in LTRACK: normal handover and tracking handover. In a normal handover the mobile equipment updates its entry in the HLR. It sends the address of the new LTRACK node to the HLR. In case of a tracking handover the mobile sends the address of the new LTRACK node to the old LTRACK node. Figure 4.2 depicts the two different handover types of LTRACK. Incoming calls always arrive to the home address of the mobile nodes, the HLR handles them. So the HLR has to locate the mobile node. It sends a request to the LTRACK node where it received the last normal handover message from. That LTRACK node either still has the mobile node connected to it, or knows a next hop LTRACK node towards the mobile node where it forwards the request. Thus, a normal handover can be followed by some tracking handovers before another normal handover takes place.