IP micro-mobility protocols

IP micro-mobility protocols Pierre Reinbold University of Namur Belgium pre@info.fundp.ac.be http://www.infonet.fundp.ac.be Olivier Bonaventure Université Catholique de Louvain (UCL), Belgium Bonaventure@info.ucl.ac.be http://www.info.ucl.ac.be March 2003 Abstract Wireless cellular networks are quickly evolving towards broadband wireless access networks, going from 2G, classical telephony networks such as Global System for Mobile communication (GSM), towards 3G and beyond, with, for example, Universal Mobile Telecommunication System (UMTS). At the same time, these networks are also moving towards all-ip networks. In this paper, we first describe the global mobility landscape for these future networks. This landscape is designed to be as generic as possible to allow us to compare several IP mobility management proposals with very different characteristics. We illustrate the utilization of this landscape with a short presentation of the mobility management in General Packet Radio Service (GPRS) networks (at the IP level) and of UMTS. Then, we point out and describe the important issues that must be addressed to manage mobile nodes. These issues include mainly handoff management, the support of Passive Connectivity and Paging, Scalability and Robustness. Within this framework, we examine Mobile IP as a first IP mobility management protocol. We present then the well-known distinction between macro-mobility and micro-mobility and the advantages of this approach. Finally, we compare seven of the recently proposed IP micro-mobility protocols: Hierarchical Mobile IP, Proactive Handoff, Fast Handoff, Telecommunication-Enhanced Mobile IP Architecture (TeleMIP), Cellular IP, HAWAII, and Edge Mobility Architecture (EMA). Keywords Mobile IP, Wireless networks, micro-mobility, macro-mobility, IP mobility 1 Introduction Today s wireless cellular networks are largely based on circuit switching technologies and are optimized to carry voice traffic. Some extensions to these networks such as GPRS [19] are currently being deployed to better support data traffic such as Internet access. Future broadband wireless networks such as the fourth generation (4G) networks for example will rely on packet switching technologies and will be entirely based on the IP protocol suite, both in the wireless and the wired parts of the network. Since IP was not designed with mobility in mind, there are several problems that need to be solved before all-ip wireless networks are deployed. A first problem to be addressed is that inside an IP network an IP address is used to identify both a node and its location. Thus, when a mobile node moves inside the network, its IP address must change. This problem has been addressed by several proposals such as Mobile IP [2]. Mobile IP offers a mechanism that allows mobile nodes to change their point of attachment (and thus their IP address) inside the network. However, when Mobile IP was designed, all-ip wireless networks were not envisionned and some of the mechanisms used by Mobile IP are not wellsuited for such networks. For example, it can be expected that voice will remain an important service in broadband wireless networks. To efficiently support voice over IP, the network will have to provide a low delay and a low delay Most of the work was done while the author was with the University of Namur. 1

jitter to the voice packets. These low delays will have to be provided even while the mobile node performs a handover operation, i.e. when it moves from one base station to another. Hence, the mobile node must be able to quickly change its IP address during the handover operation. Many approaches [4] to efficiently support mobility within IP networks have been proposed. They are often called IP Micro-mobility protocols. Sometimes designed to solve very specific problems, their heterogeneous characteristics and properties do not allow to easily obtain an accurate picture of the mobility management problems in IP networks. This paper presents a comprehensive comparison, in a global framework, of Mobile IP and seven of the main IP Micro-mobility proposals. We first present a global mobility landscape and point out the important problems to be addressed (section 2). In this first part, GPRS is presented as a case study to illustrate some important concepts. Based on this framework, we first examine Mobile IP as a Macro-mobility protocol (section 3). Then, we briefly describe and compare seven well-known IP Micro-mobility protocols in sections 4 and 5. We have chosen these protocols so that their properties offer a good overview of the most important concepts that are used in the Micro-mobility protocols. 2 A global IP mobility framework This section focuses on the presentation of a generic mobility landscape and the major issues for IP mobility to be investigated within this landscape. We use GPRS as a case study to illustrate the different concepts presented in this section. 2.1 The mobility landscape 2.1.1 A generic model All-IP networks are the expected future mobile wireless networks, relying entirely on IP, from the mobile station to the gateway towards the Internet. We call IP wireless domain 1 a large IP wireless access network managed by a AP Wireless LAN AP IP wireless access domain Inter domain backbone AP IP wireless access domain Internet Mobile node UMTS like RAN IP Layer Gateway IP router IP Wireless Point of Attachment Radio Layer Radio Network Controler Base Transceiver Station Wireless LAN Access Point RNC Figure 1: A simple model of the future IP mobility landscape 2

single administrative authority. From an IP viewpoint, we can describe such a network as a set of Wireless IP Point of Attachment (WIPPOA) connected to an IP backbone, with a gateway towards the Internet. From this point of view, the concept of WIPPOA can be understood as the smallest IP entity in the network, similar to the notion of IP subnets used in fixed IP networks. A Mobile Node (MN) can use the services offered by a domain by interacting via the radio interface with one or more WIPPOA of this domain. When a MN changes its WIPPOA, the routing of the IP packets destined to this MN must change accordingly. This situation is illustrated in figure 1. In this picture, we present a wireless domain using an UMTS-like radio interface. In such networks, the base stations, called Base Transceiver Station (), are grouped under the control of a dedicated station, the Radio Network Controller (RNC). Inside such a group(called Radio Access Network (RAN)), the mobile movements are entirely managed at the radio layer and are thus transparent to the upper layers. A RAN can thus be considered as a single WIPPOA. This is obviously only an example and the actual meaning of WIPPOA can vary between wireless domains using different radio interface technologies. In figure 1, we have also shown a domain using Wireless LAN (WLAN) as radio access points. Each of these LANs corresponds to a different WIPPOA. We also assume that each MN is attached to a Home Network (HN), a domain from which it has obtained a static IP address: its Home Address. In this context, static means that the lifetime of the address assignment is much longer than the duration of the mobile movement. We call a Foreign Network (FN) any other domain where the MN can connect. Finally, in figure 1, we have added an IP inter-domain backbone dedicated to enable the roaming between the domains, in analogy with the current GPRS networks. 2.1.2 The case of GPRS networks As an illustration, we first present the case of GPRS networks [19] 2. These networks are not all-ip network but may use IP in their backbone and they often provide IP-based data services. Moreover, GPRS offers a packet service to allow the MN to communicate with other IP networks (such as the Internet) outside the GPRS network. Many mobile operators have already deployed GPRS networks in Europe as an extension of GSM networks to provide data services to their customers. Figure 2 presents an overview of the logical architecture of a GPRS network. In such a network, the s are mostly the antennas where the MN can connect. These s are all connected to a Base Station Controller (BSC). A BSC and its depending s constitute a Base Station System (BSS). The BSSs are connected to one Serving GPRS Support Node (SGSN). The SGSNs and the Gateway GPRS Support Node (GGSN) belong to the IP-based GPRS backbone and provide IP routing functionalities 3. The communication between the SGSN and the BSS is not made over an IP network but by using dedicated network protocols. Moreover, the BSSs are grouped to improve the mobility management as we will see later. All the BSCs in a given group (called Routing Area (RA) are connected to the same SGSN. Each transmits periodically some informational messages on a dedicated channel. The mobiles use these messages to learn the identification of their current cell and the RA (among other things) so that they can discover, for example, that they have moved to another RA. The GGSN is the node used to communicate with an external Packet Data Network (PDN), such as the Internet. The GGSN is also part of the IP GPRS backbone. This backbone contains thus basically all the SGSNs, the GGSNs and a set of IP routers to interconnect them. The Home Location Register (HLR) is a node containing the GPRS subscription data and routing information. It is accessible from the GGSN and the SGSN. The GPRS network also contains other important nodes, mainly used to maintain compatibility with GSM, that are outside the scope of this paper. A major goal of a GPRS network is to allow the MN to communicate with an external PDN i.e. transmit Packet Data Protocol (PDP) Packet Data Unit (PDU), such as IP packets. Basically, a mobile can initiate a PDP session by activating a PDP Context in the network. The data forwarding across the GPRS network (between a MN and an external PDN) relies on the tunneling of packets between the GGSN and the SGSNs. The packets issued by a PDN layer. 1 We also use the words wireless domains or more simply domains to designate those networks. 2 The interested reader can find a survey of GPRS in [1] 3 GPRS does not really assume that the use of IP within the backbone is a mandatory, but we focus here on a GPRS network using IP as network 3

BSS BSC SGSN MSC/VLR SMS Entities HLR EIR Mobile node BSC GPRS signaling GPRS signaling and data transfer SGSN GPRS Backbone (IP) BSC GGSN GGSN other GPRS network Packet Data Network (IP, X.25,...) Figure 2: Overview of the GPRS logical architecture (from [19]: Figure 2) are transmitted to the GGSN. The GGSN uses a tunnel to forward them to the SGSN to which the MN is currently attached. This SGSN de-capsulates the tunneled packets and delivers them to the MN via the concerned BSS. This type of tunnel is also used when the mobile sends a packet towards an external PDN. The protocol used to create these tunnels is the GPRS Tunneling Protocol (GTP) [20]. This is illustrated in figure 3 representing a very simplified view of an IP session over a GPRS network. The focus is set on the nodes attached to the IP backbone. Mobile node IP Address : MN IP source: MN IP dest.: CN BSC SGSN IP source: SGSN IP dest.: GGSN TCP/ UDP GTP IP source: MN IP dest.: CN IP GTP TCP/UDP IP GPRS Backbone GTP tun n el GGSN IP source: MN IP dest.: CN Internet Correspondant Node IP address : CN Figure 3: GPRS data forwarding, with some layers of the transmission plane 4

This tunnel-based forwarding forces the GGSN to maintain a table with the IP address of the current SGSN serving each MN allowed to receive data packets inside the network. This table must be updated when a MN moves from one SGSN to another. This mechanism will be briefly described in the next section. From an IP point of view, we can describe the GPRS network as a set of WIPPOA, made up of a single SGSN and the depending BSSs, connected to a gateway, the GGSN, through an IP backbone. A slight difference with our simple generic model is that the WIPPOA are directly connected to other GPRS networks through dedicated IP backbones while our model assumes that all external communications pass through a gateway, even towards an inter-domain backbone. In GPRS networks, this feature is used, for example, when a mobile is connected to a different network than its home network. In this case, the serving SGSN establishes a direct GTP tunnel towards the GGSN of the home network. 2.2 Major mobility issues We now define some important issues for the mobility management in our mobility landscape. The comparison will be made with respect to these topics on the basis of the following performance criteria : latency, amount of control traffic, robustness and scalability. 2.2.1 Handover management From the viewpoint of IP, the handover concerns the management of the changes of WIPPOA of the mobiles during their moves. The handover management is obviously a major issue in mobility management since a MN can experience several handovers during a single session as in current cellular telephone networks. 2.2.2 Different types of connectivity Mobile devices have a very limited power capacity and their batteries must be spared by reducing the mobiles transmissions to the minimum required. This is a common problem in mobile telephony. However, in order to forward the traffic to a mobile, the network must maintain some information on its geographical location. When the mobile transmits packets, the handover management ensures that the network always knows the current location of the MN i.e. the cell where it is located. This procedure must obviously be fast and efficient to ensure that the mobile experiences little perturbation during handover. We call this type of mobility management the active connectivity mode, because the MN is actively transmitting data and not only signaling traffic inside the network. Unfortunately, even if a mobile is not transmitting any data, the network must maintain information about its location. If a MN is not transmitting and changes its WIPPOA, it will be impossible to forward an incoming packet destined to it since the network does not know where the mobile is located. In most cases, the active handover management procedures (in active connectivity mode) generate a lot of control traffic because each change of cell must be signaled and managed. This is very power consuming for the mobiles and could be a burden for the network. In order to conserve the battery power of the mobiles and to reduce the signaling traffic inside the network, wireless networks often define another working mode for the mobiles that do not currently transmit data. We call this mode the passive connectivity mode. In most cases, it implies the construction of a paging architecture. This architecture consists in dividing the network in distinct geographical areas that we call paging areas. As an example, we will see in section 2.2.4, that the passive and active connectivity mode are translated in GPRS networks in the states machine of the mobile with stand-by state and ready state. We will also see that the paging areas are called Routing Areas in these networks. The mobility management proposals use different names for all these concepts. We have chosen to define passive connectivity and active connectivity, as used [5] for example, in order to use a common terminology when comparing several micro-mobility proposals. When the mobile has no data to transmit it switches to passive connectivity mode and only issues a beacon when changing of paging area. This implies that the network only knows the approximate location (the current paging area) of the mobile. Any incoming data destined to a passive mobile forces the network to utilize a paging procedure to 5

find its precise location (the cell where it is located inside its paging area). At the end of this procedure, the network can deliver the data directly to this cell. In most cases, the mobile switches then to active connectivity mode so that the paging procedure is made only once for a data flow. On the other hand, to send data while in passive connectivity mode, the MN must first inform the network of its current location and move to the active connectivity mode. Enabling passive connectivity support via a paging architecture can be an additionnal burden for the network (at least in terms of control traffic). It is important to realize that this architecture is used to support only incoming data packets. This solution must thus be carefully considered with respect to efficiency concerns, such as the ratio of incoming and outgoing communications or the number of handovers experienced by the mobiles. 2.2.3 Intra-domain traffic Intra-domain traffic is an important part of the current traffic in wireless networks such as GSM/GPRS (the Short Messages Service (SMS) for example). We can expect that the mobile users of future wireless networks will also communicate a lot between mobiles connected to the same domain. This type of traffic must thus be efficiently supported. 2.2.4 Mobility issues and GPRS As an example, we briefly describe how these mobility issues are supported in the GPRS networks. Mobility Management in figure 4. From a GPRS point of view 4, a mobile node can be described by the state machine illustrated Mobile not Reachable IDLE GPRS Detach GPRS Detach GPRS Attach Mobile Reachable The network only knows its current Routing Area STAND BY READY timer expiry or Force to STANDBY PDU transmission (data or signalling) READY Mobile Reachable The network knows its current Cell Figure 4: The GPRS Mobility Management state machine, MN side, from [19]: Figure 13 In the IDLE state, the MN is not reachable: it is not attached to the GPRS network. In the STANDBY state, the mobile is attached to the network and involved in the Mobility Management. It performs a routing area update when changing of RA. Each time it detects a change of RA it issues a special message to inform the network of its current location. In this way, the network only knows the RA where the MN is currently attached. To establish a data flow towards the mobile, it must first find its location inside this RA. The SGSN in charge of the RA will send a request to all the BSSs of the area, so that each will broadcast a request message on a dedicated channel. All mobiles listen to this channel and the concerned mobile will send a reply to the GGSN. 4 We focus here on the IP side of the GPRS network. 6

The state of the MN is changed to READY when it sends data or signaling information (such as a response to request as above). The mobile can also move to IDLE state by initiating a GPRS Detach procedure. After the expiration of the Mobile Reachable Timer, the SGSN assumes that a mobile in STANDBY state is now in the IDLE state. In the READY state, the MN is reachable in this state, attached to the network and involved in the Mobility Management. It informs the network each time it changes of cell (by a cell update procedure or a RA update procedure). The mobile moves to STANDBY state after the expiration of the Ready Timer. This timer must be carefully tuned. It must be sufficiently large to reduce the number of paging procedures that can be triggered by bursty traffic (such as IP traffic). If each burst triggers a paging procedure, the signaling traffic will become very high. The SGSN can also force a mobile to move to the STANDBY state. The mobile may move to the IDLE state by initiating a GPRS Detach procedure. In a GPRS network, an IP handover is called packet rerouting. It implies that the IP routing (in the backbone) changes for the concerned mobile. This occurs only when the MN changes of SGSN. In this case, the network must create a new GTP tunnel between the GGSN and the new SGSN. As all s in a RA are connected to the same SGSN, each packet rerouting implies an inter-sgsn Routing Area Update. We only describe this kind of update procedure as it is the most important in the perspective of an IP mobility management. It is described in the figure 5 but we focus here only on the IP mobility management in the case of a successful registration. 1. Routing Area Update Request 2. SGSN Context Request 2. SGSN Context Response 3. Security Functions 4. SGSN Context Acknowledge 5. Forward waiting packets 6. Update PDP Context Request 6. Update PDP Context Response 7. Update Location 9. Insert Subscriber Data 9. Insert Subscriber Data Ack 8. Cancel Location 8. Cancel Location Ack 10. Update Location Ack 11. Routing Area Update Accept 12. Routing Area Update Completed Figure 5: Inter SGSN Routing Area Update Procedure, from [19]: Figure 28 7

The procedure begins when the MN is connected to the new inside the new RA. The mobile sends an Update RA Request to the new SGSN. Upon reception of this request, the new SGSN sends a SGSN Context Request to the old SGSN. This has two main goals: allowing the new SGSN to authenticate the MN with information provided by the old SGSN via the SGSN Context Response; informing the old SGSN of the address of the new one so that it may forward any packet arrived for the MN during the Routing Area Update. The new SGSN sends an acknowledgment to the old one to trigger the forwarding of waiting packets. The new SGSN must then inform the GGSN of the new WIPPOA of the mobile. This is done via an Update PDP context request/response exchange. At this time, the GGSN has updated the SGSN address for the mobile and the network knows all the required information to correctly forward the packets sent or received by the mobile. The new SGSN then contacts the HLR to update its location data for the MN. The HLR sends a Cancel Location to the old SGSN so that it removes the PDP context associated with the mobile. A timer is associated with these contexts in the SGSN to remove outdated PDP contexts. The old SGSN acknowledges the messages and the HLR can send Insert Subscriber Data to the new SGSN. Upon the reception of such a message, the new SGSN needs to validate the location of the MN in the new RA. After the last checking and if they are successful, the new SGSN builds a new context for the mobile and sends an acknowledgment to the HLR. After a last ack from the HLR, in the case of a successful update, the new SGSN sends a Routing Area Update Accept to the MN. The mobile acknowledges it to complete the procedure. Intra-domain traffic We focus here on an exchange of IP packets between two MNs connected to a GPRS network. Once they have activated an IP PDP context, the MNs can exchange IP packets. Figure 6 illustrates this exchange of packets. We can see that all packets pass through the GGSN. This will occur even if the two mobiles are connected to the same SGSN. This is obviously not an optimal solution from the viewpoint of routing efficiency. We will see that this type of intra-domain routing, often called triangular routing, also occurs with Mobile IP. GPRS within our framework On the basis of the GPRS mobility management, we can now describe the GPRS network within our framework. The MN can experience different kinds of handover. When it moves between s that are all connected to the same SGSN (even through different BSCs), it experiences then what we will call a radio handover as opposed to an IP handover (packet rerouting). These radio handovers are entirely managed outside the IP backbone and do not influence the IP routing. For this reason, a WIPPOA in a GPRS network is composed of a SGSN and all the related BSCs with their s. In currently deployed networks, such a WIPPOA can represent a large geographical area for a single point of attachment. We can also describe the GPRS MN states machine by using the connectivity modes of framework. When the GPRS MN is in STANDBY state, it can be described as in passive connectivity mode within our framework (we use here exactly the same definition as [5]). On the other hand, the MN in READY state will be labeled as in active connectivity mode within our framework. In the IDLE state, the MN has obviously no connectivity at all. 2.3 A glimpse at 3G networks In this section, we briefly discuss UMTS networks, as currently being standardized by 3GPP [32], to provide an overview of the mobility management within 3G networks [13]. 2.3.1 A brief overview of UMTS At first glance, UMTS re-uses the GPRS mobility management procedures [21, 23, 22]. The main difference is that UMTS, and in particular UMTS releases 4 and 5, relies more intensively on IP than GPRS. UMTS networks have two 8

IP source: MN1! IP dest.: MN2 MN BSC SGSN GPRS Backbone GTP Tunnel 1 GGSN SGSN GTP Tunnel 2 BSC MN Figure 6: Intra-domain routing in GPRS network distinct parts : Circuit Switched (CS) and Packet Switched (PS). Inside the UMTS Core Network (UCN), these parts are called domains. IPis used in the PS domain of the UMTS network and is equivalent to the IP backbone used in GPRS networks. However IP is also used to connect the radio access network (Universal Terrestrial Radio Access Network (UTRAN)) and the PS domain of the UCN. As shown in figure 7, UMTS re-uses several architectural principles of GPRS. Inside the UTRAN, we find two types of entities: UMTS Radio Network Controller (URNC) and Node B (NB). They are the UMTS counterparts of the GPRS BSC and respectively. An URNC and its depending NBs constitute a Radio Network Subsystem (RNS), equivalent to a GPRS BSS. An important difference between GPRS and UMTS is the mobility management. While GPRS uses a state-machine with three states to describe the MNs, UMTS defines the concept of radio level connection (RRC connection). The UMTS terminals can thus be described (roughly) as being in two states: connected at the radio layer or not. Moreover, the radio layer connection can be in four different states corresponding to its transmission activity. When the MN is not connected, the network behaviour is very close to the GPRS mobility management in the STANDBY state. Similar procedures are used, like RA update and paging requests/reply for example. When the RRC connection is established, from an IP mobility point of view, we have a kind of two level hierarchy inside the UMTS IP network. At the first level, inside the UCN, the packets sent by the users are managed as in GPRS by using GTP tunnels between the UMTS GGSN and the UMTS SGSNs. At the second level, between the UCN and the UTRAN, GTP tunnels are also used to enable the communication between the SGSNs and the URNCs. These tunnels are dynamically established and removed by a new procedure called SRNS relocation [21] to manage changes of Serving RNS for a MN. The Serving RNS is the RNS where the MN has established the RRC connection. This RNS may be different from the RNS where the MN is actually located (called Drift RNS), inside a given RA. 9

$ Circuit Switched Domain MSC, VLR, GMSC,... SGSN EIR HLR GTP tunnel UMTS Core Network GGSN SGSN # Correspondant Node Packet " Data Network (IP, X.25,...) Node B GTP tunnel RNC Node B Node B Mobile node RNS UMTS IP Network RNC Node B Node B UTRAN UMTS signaling UMTS signaling and data transfer Figure 7: Overview of the UMTS logical architecture and data forwarding In this case, the packets must obviously be forwarded from the Serving RNS to the Drift RNS. The SRNS relocation allows the UCN to forward the packets directly to the RNS where the MN is located. The IP mobility management procedures used inside UMTS networks are specific cases of the SRNS relocation procedure. In case of a change of RA for example, this procedure occurs across the UCN and is performed between the current SGSN, the GGSN and the HLR. On the other hand, the SNRS relocation is limited to the SGSN when the mobile does not change of RA. These two levels define the IP mobility management hierarchy inside UMTS. In figure 7 we can see the two GTP tunnels used to forward the mobile data packets across this hierarchy. 2.3.2 UMTS in our framework Within our framework, a UMTS WIPPOA is thus composed of the Serving RNS (instead of the GGSN and all depending BSS with GPRS). We can describe a MN without an RRC connection as being in a passive connectivity mode. In the same way, a mobile with an RRC connection to the UMTS network can be considered in an active connectivity mode. However, UMTS defines four different states for the connection, with respect to the activity of the mobile. Such a precision can not be described within our framework. 10

' ( ) * 3 Mobile IP and all-ip wireless networks In order to introduce the comparison of the micro-mobility protocols, we first present a brief overview of Mobile IP [2] and its major drawbacks that have led to the definition of the micro-mobility approach. 3.1 Mobile IP Mobile IP is probably the most widely known mobility management proposal. Its simplicity and scalability give it a growing success. Mobile IP is described in [2] (a good review paper can be found in [31]). Several extensions and enhancements are described in [28, 29, 35, 9]. Here, we discuss the principles of Mobile IP and ignore the differences between the IPv6 and IPv4 versions. To allow a mobile IP node to change its WIPPOA, Mobile IP defines two types of Mobility Agent (MA) : the Home Agent (HA) and the Foreign Agent (FA). The HA is located inside the home network of each mobile and a FA is located inside each foreign network where a MN can connect. Mobile IP uses a couple of addresses to manage user s movements. Each time the MN changes its WIPPOA, it obtains a temporary address called Care Of Address (COA) from a FA directly connected to this WIPPOA. The presence of the FA in a particular subnet can be detected via FA advertisement messages that are extensions to Internet Control Message Protocol (ICMP) router advertisement messages [10]. Those messages are broadcasted at regular time intervals by the FA. The MN can also send advertisement messages to trigger a FA to transmit its advertisement message. Each time a FA delivers a COA to a new MN, it must insert a binding for this mobile in a dedicated table called its visitor s list. Once the MN has obtained its new COA, it must inform its HA of this new address by using the registration process. As soon as the HA is aware of the MN s current COA, it will intercept the packets received in the home network for the MN. The HA will then tunnel those packets to the FA. Upon reception of encapsulated packets, the FA will deliver the original packet to concerned the mobile node. This is illustrated in figure 8 Correspondant Node IP address : CN ' IP source: CN IP dest.: MN R From the Correspondent to Mobile IP within IP encapsulation From the Mobile to Correspondent & % Home Network R IP source: MN ' IP dest.: CN Foreign ' Agent IP Address : FA % Foreign Network Internet R R ' IP source: HA IP dest.: COA IP source: CN ' IP dest.: MN Mobile node IP Address : MN Care of address: COA Home Agent IP Address : HA Figure 8: Basic working of Mobile IP 11

3.2 The micro-mobility approach Mobile IP suffers from several well-known weaknesses that have led to the macro/micro-mobility approach. In this section, we review some of these weaknesses to introduce this approach. We also point out several important properties shared by the micro-mobility proposals considered in this paper. In Mobile IP, the basic mobility management procedure is composed of two parts: the movement detection by the MN and the registration to the HA. Every time the mobile changes its WIPPOA, these two steps must be accomplished to allow the MN to continue to receive packets. However, it is the MN that initiates the process by sending a registration request once it has detected that it moved from one network to another and has obtained a new COA. This introduces two causes of latency: move detection latency and registration latency. The move detection latency is the time required by the MN to detect that it has changed of WIPPOA. It can be large since the move detection mechanisms in Mobile IP are based on either the expiration of the lifetime of FA agent advertisements 5 or on the comparison of the address prefix of two different agent advertisements. The registration latency is the required time to complete the registration with the HA. As this HA can be located anywhere on the Internet, this process can take a long time and sometimes be impossible to complete. Those latencies, due to the properties of Mobile IP, introduce a delay before the packets destined to the MN can be routed correctly towards its new location. During this time, the mobile is already connected to its new WIPPOA but the packets that are sent to it are routed to its old WIPPOA. These packets are thus lost for the MN. In the case of a quickly moving mobile which changes of network rapidly, the registration process will become totally inefficient. Moreover, this mechanism produces a lot of control traffic inside the local domain and across the Internet. The micro-mobility approach tries to reduce the latency of the handover management. This approach does not always reduce the control traffic, but it allows to reduce the number of network stations that process the control packets by restricting the propagation of those packets to a smaller set of stations. The micro-mobility approach implies the utilization of two different protocols to manage the mobility: Mobile IP is used to manage the movements of the MN between distant wireless domains and across the Internet (macro-mobility), A micro-mobility protocol is used to manage the movement of the MN inside each wireless domain. A micro-mobility protocol behaves as follows. The MN obtains a local COA when it connects to a domain. This COA remains valid while the MN stays in this domain and the mobile will thus perform only one home registration (registration with the HA) when it first connects to the domain. The users movements inside the domain are managed by a micro-mobility protocol. This is transparent to the HA and the rest of the Internet. In fact, for the HA, each wireless domain is considered as a single Mobile IP subnet. Latency and control traffic across the network are thus extremely reduced. This is the main reason to adopt a micro-mobility approach. On this basis, each micro-mobility proposal aims at reducing the move detection latency (by an interaction with the radio layer for example) and at optimizing the handover management inside a domain. 4 Comparison of Micro-mobility protocols This section presents a short description of the main elements of the different micro-mobility protocols that we have selected and, for each protocol, an evaluation against the criteria defined within our framework. 4.1 Comparison criteria We define here the four criteria selected with respect to our framework and the different parameters used to compare and evaluate the micro-mobility proposals. 5 This is the lifetime indicated in the ICMP router advertisement 12

S R o n P P T V U W X Y Z [ \ ] ^ _ ` a b c d 4.1.1 Handoff The handoff management is the most important issue in the mobility management. We investigate it on the basis of the simple network model shown in figure 9 with respect to: handoff management parameters: the interaction with the radio layer, initiator of the handover management mechanism, use of traffic bicasting, etc., handoff latency: the time needed to complete the handoff inside the network, potential packet losses: the amount of lost packets during the handoff, involved stations: the number of MAs that must update their routing data or process messages during the handover. For this comparison, we assume that +-,/.1032 is the average number of hops between a MN and the gateway. The delay between these two hosts is 4,/.10326587:9<;. Similarly, +-=?> 2A@ is the number of hops between a MN and its former WIPPOA (delay: 4?=/> 2A@B5C7:9:; ). 4?DE>GF?HAH is the average delay in 5C7:9:; between the MN and the so-called crossover node for a given handoff. This node is the first common node located on both the path between the new WIPPOA and the old WIPPOA and on the path between the new WIPPOA and the gateway. In the case of FA based mobility management (see section 4.2.1), these concepts must be understood in terms of FA. In general, we can assume that 4,/.1032JI 4?=?> 2A@KI 4?DE>?F?HAH. 4?LNM is the average time needed to reach the HA with the classical Mobile IP registration mechanism. Base s e tati f ons g cove h r i ag j k lrm ea e a New Base IP POA Station Crossover Station Node tcross tprev tgate OMobile s movement QGateway B a s e s ta ti on Previous Base IP POA Station s cove Figure 9: A simple model to compare handoff mechanisms ra Internet When investigating the performance of handover mechanisms in micro-mobility protocols, we must consider the important issue of move detection. We have already seen that the micro-mobility approach reduces the registration latency since most of the registrations are restricted to the current domain. However, the detection of the occurrence of a handoff is another important source of delay for real-time applications. As the IP handover management occurs after the movement detection, this detection must be as fast as possible. In other words, any IP handover management mechanism is useless if the movements of the MN are detected too late and packets have already been lost. In Mobile IP, the movement detection is performed by using two algorithms described in [2]. These algorithms are based on the ICMP router discovery messages. Handoff is detected when receiving a Mobility Agent Advertisement with a source address located in another network (beginning with a different prefix) or when the lifetime of the last Mobility Agent Advertisement received expires. With the first algorithm, the detection occurs, on average, after the time between two Agent Advertisement (twice this time in worst case). With the second algorithm, it occurs after the lifetime of the Agent Advertisement. The values of these parameters should be adapted to the local network (their default values are 30 min. for the lifetime and 7-10 min. for the time interval between two Agent Advertisement [10]). We will call this latency 4?prq =. In the case of protocols relying on an interaction with the radio layer, we call st4?0 > qu,/,/2 > the time between the reception of the radio trigger (ex. Strong Handoff Radio Trigger (SHRT)) and the actual radio handoff i.e. the moment g e a r ea 13

x x x when the radio link between the MN and its old WIPPOA is removed. This time interval obviously depends on the radio technology, the load, the local topology of the network, the MN movements,... In figure 9, st4?0 > qv,/,/2 > may represent the time that the mobile crosses an overlapping area, going from point w to point y, if, for example, the radio trigger is sent when the mobile is at point w. For each proposal, we have defined the uncertainty time. During this time interval, after the radio link with the old WIPPOA is deleted, the packets destined to the MN may be lost or incorrectly routed in the network. This parameter is very important since it reflects the efficiency of the handoff management mechanism with respect to packet losses. In our comparison, we do not take into account the time required by the MN to reach the stations on the wireless interface. This time, which can be long, is not relevant for our comparison. Moreover, we also neglect the time required to transfer packets inside a RAN for the FA based mobility management proposals. We consider that this is a layer 2 characteristic. However, it can be large in the case of protocols assuming interaction between layers since a RAN can cover a very large geographic area. 4.1.2 Passive connectivity and Paging We have described the support of the passive connectivity in section 2.2. Only a few proposals define a support for this feature by using a paging architecture. For these protocols, we evaluate the algorithm used to perform the paging with respect to efficiency and network load. 4.1.3 Intra-network traffic Intra-network traffic corresponds to the packets exchanged between MNs connected to the same wireless network. This kind of communication is a large part of today s GSM communications and we can expect that it will remain an important class of traffic in future wireless networks. The efficient support of this type of traffic is thus an important concern. 4.1.4 Scalability and robustness Current mobile networks support millions of connected users communicating at the same time. We can expect that future large wireless access networks will have the same constrains in terms of users load. For example, a commercial router acting as GGSN in a GPRS network is able to manage 90,000 simultaneous user contexts [8]. These facts are to be related to the increasing load of today s Internet routers: routing tables containing a few hundreds of thousands entries have become a performance problem. We evaluate the different proposals with respect to their scalability and the stations requirements within the network. 4.2 Selected IP Micro-mobility protocols In this section, we present the different micro-mobility proposals and we evaluate them with respect to our framework. Each subsection is organized as follow: a short description of the protocol, followed by the evaluation of this protocol against our criteria. We group four proposals under the label FA based mobility management proposals. These are extensions to Mobile IP for the micro-mobility management and the interaction between different FAs in the network is the basic mechanism of the mobility management for these proposals. These proposals are Hierarchical Mobile IP [17] and its extension for the regional paging [18], Fast Handoff [12], Proactive Handoff [3] (see also [11]) and finally TeleMIP [14] that is more an architecture than a real protocol. We also include in the comparison Cellular IP [5], HAWAII [33, 34] and EMA [25, 24]. All these protocols are the result of the standardization process within the IETF Mobile IP [15] and the Seamoby [16] working groups. A comparison of the handoff performance of Cellular IP, HAWAII and Hierarchical Mobile IP can be found in [7] and a good taxonomy of the different micro-mobility proposals may be found in [4]. 14

4.2.1 FA based mobility management proposals Several micro-mobility proposals manage the user mobility on the basis of interactions between FAs. Hierarchical Mobile IP [17], is an extension to Mobile IP that supports a hierarchy of FAs between the MN and the HA. Several improvements to Hierarchical Mobile IP, including a paging mechanism, have been proposed in [18]. Fast Handoff [12] and Proactive Handoff [3] are two similar proposals based on Hierarchical Mobile IP with improved handoff mechanisms. TeleMIP [14] adds some load balancing features to the basic principles of Hierarchical Mobile IP. A basic network model for these proposals is shown in the figure 10 (left). Here we have a set of WIPPOAs and each is associated with a dedicated FA. The FAs are connected to a so-called Gateway Foreign Agent (GFA). This network model is used by Hierarchical Mobile IP and Proactive Handoff. TeleMIP uses the same model but allows the FAs to be connected to several GFAs. Figure 10 (right) shows a slightly more complex architecture with a multi-level hierarchy of FAs between the GFA and the leaf FAs. Each FA in the hierarchy may be in charge of a WIPPOA. This type of network is described in the appendix B of [17] and is used to support Fast Handoff. zfa GFA zfa FA GFA zfa zfa FA FA zfa FA Figure 10: Network models for FA based mobility management A. Short descriptions of the proposals Hierarchical Mobile IP Hierarchical Mobile IP [17] is a natural extension to Mobile IP to efficiently support the micro-mobility approach. After the first connection of a MN to a domain and its home registration with the address of the GFA as COA, the MN will perform Regional Registrations only. Those registrations are sent by the mobile to the GFA each time it changes of FA (i.e. of WIPPOA). A regional registration contains the new local COA of the MN: the address that can be used by the GFA to reach the MN while it remains connected to the same FA. This address can be either a co-located address or the FA address. The routing of IP packets with Hierarchical Mobile IP is then very simple. A packet destined to the MN is first intercepted by the HA and tunneled to the GFA. Then, the GFA de-capsulates and re-tunnels it towards the current local COA of the MN. Hierarchical Mobile IP also supports a multi-level hierarchy of FAs between the leaf WIPPOA and the GFA. Each FA in the hierarchy must maintain a binding in its visitor s list for each MN connected to a WIPPOA lower in the hierarchy. These bindings are established and refreshed by regular registration requests and replies that the mobiles exchange in the network. The regional registrations sent by a MN are only forwarded to the first FA that already has a binding for this MN. The upper levels of the hierarchy are not aware of the details of the mobiles movements since they do not need to change their binding. In this way, the handoff management is limited to a very small number of nodes. In addition, [18] has introduced a paging support for Hierarchical Mobile IP. It relies on paging areas that are sub-trees of a single hierarchy (all the stations belonging the the same paging area belong to the same sub-tree). In each of these areas, the root of the sub-tree is called a Paging Foreign Agent (PFA). It maintains a specific visitor s list with an idle flag set for each mobile in passive connectivity mode currently located inside this area. The PFA is in 15

charge of the entire paging process by performing the paging request and managing the incoming packets destined to passive MNs. This paging mechanism is not included in the other FA based mobility protocols discussed later in this section. The architecture of Hierarchical Mobile IP is similar to the architecture of GPRS, especially if the Hierarchical Mobile IP network has only a two levels hierarchy. The GFA is similar to the GGSN and the leaf FAs represent WIPPOAs, similar to the SGSNs. Both protocols use tunnels to forward the traffic inside the network. Their main difference lies in their respective utilization of IP. GPRS uses IP only inside the backbone. Hierarchical Mobile IP assumes an all-ip network. This difference is illustrated by the size of the WIPPOAs. In GPRS a WIPPOA corresponds to a single SGSN. This implies a very large size since the BSCs and their s connected to a single SGSN can cover a wide area. In Hierarchical Mobile IP, IP is pushed to the last level before the s and an WIPPOA is assumed to cover a smaller geographical area. Moreover, the way that the paging architecture is implemented in these two proposals is directly linked to their own architecture. In GPRS, the belonging to the same RA must be connected to the same SGSN. Hence, a single WIPPOA potentially includes several RA. On the other hand, in Hierarchical Mobile IP, the paging areas are sub-trees of the main hierarchy and thus include several WIPPOA. Another important difference between these two protocols is that GPRS includes security and accounting functionalities. To implement these features, the GPRS network uses the HLR, an entity that does not exists in Hierarchical Mobile IP, and a more complex (and also longer) handover management procedure. Fast Handoff Fast Handoff [12] re-uses the architecture and principles of Hierarchical Mobile IP and addresses two remaining problems of this proposal. These are mainly the need for a fast handoff management for real-time applications and the presence of triangular routing inside the domain. In the previous section, we have seen that Hierarchical Mobile IP does not improve the Mobile IP movement detection: it relies on the ICMP Router Advertisement messages used by Mobile IP. Fast Handoff assumes the possibility of an interaction with the radio layer to anticipate the handoff and allows the MN to perform its registration with a new FA via the old FA before the handoff actually occurs. We assume here that the IP layer receives these handoff events as triggers from the radio layer. We use these triggers as a simple way of representing the interaction between the radio and the IP layers during the handover. In most cases, the radio layer is constantly doing power measurement on the signals received from its peers (for example during the cell selection procedure). On the basis of these measurements, it is possible to evaluate the signal quality for a particular node and to detect that a handover is at the point to append. When the handover occurs, the radio layer informs the IP layer with a dedicated trigger. In Fast Handoff, these triggers are designed to inform the IP layer of the imminence of a handoff by providing the next WIPPOA of the MN (the IP address of the new FA). We call this interaction with the radio interface: SHRT, as it contains the new WIPPOA of the MN. These SHRT can be received by any of the nodes involved in a handover process: the MN as well as the WIPPOAs. Moreover, they can receive these triggers at different times during the handover. On this basis, the protocol contains two mechanisms to perform the handoff with respect to the capabilities of the radio layer when a mobile is able to communicate with more than one base station. A global overview of these mechanisms is shown in the left side of figure 11. In this figure, we can see that the first step is an interaction between the radio and the IP layer: the reception of the SHRT by the new FA. This FA then sends an agent advertisement to the MN via the old FA. Finally, the MN registers with the new FA using its current radio link with the old FA. It is also possible to use the bicasting [2] capabilities of Mobile IP with simultaneous bindings to reduce the possibility of packet losses. Fast Handoff is the only proposal addressing the problem of triangular routing inside the domain (cf. 2.2.4, third paragraph). This is done by using the information found in the visitor s list. When a FA receives a non-encapsulated packet coming from a MN, it consults its visitor s list to see whether it contains an entry for the destination address. If it contains one, the FA can directly send the packet to this address. Otherwise, it forwards the packet as a normal Mobile IP packet. Figure 12 illustrates the triangular routing problem. In the left part of the figure, the intra-domain triangular routing is shown as it appears with Mobile IP. The IP packets sent by MN1 are routed to the HA of MN2 and return then to the domain where the mobiles are located. In the middle, we can see the partial solution that is offered by the route optimization extension: MN1 knows the COA of MN2. When this address belongs to the FA, the 16