Mobility Management 
in IETF and GPRS Specifications

 

15th May 2000

Sami Ala-Luukko
Department of Computer Science and Engineering
Helsinki University of Technology
Sami.Ala-Luukko@sonera.com

Abstract

IP based service technologies are becoming increasingly important in wireless communications. Cellular networks will be used as an access method to the Internet and other IP-based networks. There have been several attempts to optimize cellular networks for the packet data. One of them is GPRS (General Packet Radio Service) which is an enhancement to the existing GSM system. This report studies mobility management methods applied in GPRS and other wireless, non cellular networks. On macro mobility level, GPRS tunneling protocol and Mobile IP are compared. On micro mobility level, Cellular IP is compared to the methods applied in GPRS. Where needed within the scope of this study, characteristics of Wireless LAN radio access (IEEE 802.11) is compared to the characteristics of GPRS radio access.

Contents

1. Introduction

2. Macro Mobility Issues

2.1. Mobility Management in Mobile IP and GPRS Tunneling Protocol
2.1.1. Tunneling Methods to Support Mobility
2.1.2. Tunneling End-Point
2.1.3. Default Router Functionality and Connectivity to External Networks
2.1.4. Registration and Tunnel Establishment
2.2. Convergence between Mobile IP and GPRS Tunneling Protocol
2.2.1. IP Host Roaming and Accessing External Networks
2.2.2. PPP Support
2.2.3. Dynamic IP Address Allocation
2.2.4. Functions on Cellular Network Operator Level
2.2.5. Convergence Status in GPRS/UMTS Release '99 Standards
3. Micro Mobility Issues
3.1. Mobility Management States
3.2. Paging Area Concept
3.3. Lower Layer Issues
3.4. Packet Routing
3.5. Handoff Decision


5. Conclusions

          5.1. Macro Mobility Management
          5.2. Micro Mobility Management
 
 

References
 
 

1. Introduction

Telecom and Internet technologies have had different evolution paths. The telephone made its public debut during year 1876 [1]. After that, telecom networks were built and optimized to transport circuit switched telephony traffic. Transmission was first based on analogue and later on digital transmission technology. Compared to telephony, the Internet is a relatively new innovation. Internet is stated to be an outcome of the initial research objectives articulated by the Advanced Research Projects Agency (ARPA) in the late 1960's [1]. There has also been several different technologies which have profoundly affected the evolution of the Internet. These include IP router technology, Unix operating system, Ethernet local area network technology and Personal Computers [1]. Nowadays, the Internet is a global network which consists of various commercial and non-commercial IP backbones. An open, IP level interconnection between these IP backbones forms the concept of one, global Internet.

One key component in telecom networks has been an exchange which handles, e.g., call control, call switching, charging and call routing. On another hand, IP router has been one key building block of the Internet. The functionality of a typical IP router is very different from a telecom exchange. A basic IP packet forwarding applied in the Internet does not require a concept of end-to-end connection from router's point of view. The principles of resource reservation are also very different in the Internet and in the telecommunications networks. The Internet has typically worked by using a best effort traffic model. Every packet is treated (forwarded or discarded) equally. This has been a very simple and efficient model and several arguments has been stated against any need for a more complicated system [2]. Telecom networks, on another hand, have traditionally made an explicit reservation of transmission resources for each call.

Unfortunately, the existing telecom exchanges are not very capable of supporting bursty and high speed data traffic. However, from commercial point of view, a reuse of existing fixed access technology is very important. E.g., asymmetric digital subscriber line (ADSL) is reusing the access infrastructure of the existing public switched telecommunications network (PSTN). The existing telecom exchanges do not have any role in packet forwarding from a ADSL subscriber to an IP network (or vise versa). There have also been several attempts optimize cellular networks for packet data. One such example is General Packet Radio Service (GPRS), which is an enhancement to the existing GSM network (Global System for Mobile Communications) [3].

IETF Mobile IP  working group is aiming to understand data service in cellular systems. However, IETF is not focused only on cellular networks. One must also remember that not all wireless networks are cellular networks. If we compare the existing cellular technologies, e.g., to Wireless LANs (WLAN), we can point several differences. One important difference is that the existing cellular technologies are commonly built for wide area coverage. Other typical differences are the principles of resource reservation: cellular networks carry out explicit resource reservation over the radio interface. In this sense, cellular networks can be seen as a wireless extension to the PTSN. WLANs, on another hand, are reusing many of the principles applied in fixed local area network technologies (e.g. Ethernet).

A clear drawback of the existing cellular networks is a limited data rate: for example the most commonly applied GSM data service is only a 9.6 kbps circuit switched connection. The coverage offered by WLAN technology is limited, but the bit rates are higher. A typical existing WLAN implementation (compliant to IEEE 802.11) offers 2 megabits per second in a shared access manner [4].

Figure 1 present the architecture studied in this paper. This report studies mobility management methods applied in GPRS and other wireless, non cellular networks. On macro mobility level, GPRS tunneling protocol and Mobile IP are compared. On micro mobility level, Cellular IP is compared to the methods applied in GPRS. Where needed within the scope of this study, characteristics of Wireless LAN radio access (IEEE 802.11) is compared to the characteristics of GPRS radio access.
 
 

Figure 1. Architecture studied in this paper.



2. Macro Mobility Issues

Mobile IP work in IETF applies terms macro mobility and micro mobility [5]. The basic specification for Mobile IP (RFC 2002) is giving tools for the macro mobility management. However, the basic Mobile IP is not that well suited for micro mobility [6]. A typical example of micro mobility is a handoff amongst neighbor wireless transceivers, each of which is covering only a very small geographic area.

In GPRS (as in most cellular systems), network is divided into two parts: core network and radio access network. In GSM, radio access network is called base station subsystem. In core network, from packet forwarding point of view, GPRS specific functionality is divided into two network elements. From network topology point of view, Serving GPRS Support Node (SGSN) has direct connectivity to radio access network. Gateway GPRS Support Node (GGSN) has direct connectivity to the external IP networks (e.g. Internet and intranets). GPRS core network in general carries out many similar macro mobility issues which are supported by Mobile IP. This will be discussed in Section 2.1. Micro mobility issues will be further discussed in Section 3.

2.1. Mobility Management in Mobile IP and GPRS Tunneling Protocol

There is some difference in standardization models applied for GPRS and Mobile IP. GPRS is an entire mobile network system where architectural descriptions and interfaces between network elements are standardized. Mobile IP, on another hand, is a protocol applied in the IP environment to support a mobility of IP hosts. In IETF, it has been typical to standardize more or less independent protocols. From architectural point of view the IETF model potentially allows more flexibility. Especially in backbone networks, IETF protocols have become more and more dominant. From cellular terminal implementation point of view, a specification of entire network system and thus a complete radio interface might have been beneficial for interoperability purposes.

GPRS tunneling protocol (GTP) carries out many similar functions as Mobile IP. The following sections point out some similarities and differences when Mobile IP and GTP are compared from mobility management point of view.

2.1.1. Tunneling Methods to Support Mobility

As it is widely known, IP version 4 assumes that node's IP address uniquely identifies the node's point of attachment to the IP network [6]. To support the mobility of IP hosts, both Mobile IP and GTP apply tunneling to support packet forwarding over those routers which do not have mobility specific functionality. The Internet routers do not have to be aware of the IP host mobility. Also from performance point of view it is not always feasible that, e.g., high speed backbone routers are performing mobility management specific functions. When away from home, Mobile IP uses protocol tunneling to hide a mobile node's home address from intervening routers between home network and host's current location [6]. In GPRS, user data is transferred transparently between the mobile station and the external data networks using encapsulation and tunneling. GTP (which is primarily applied between GGSN and SGSN) transfers end user data through tunnels [3].

There are different methods of implementing the tunnel over intermediate routers. In Mobile IP, both home agent and foreign agent must support tunneling datagrams using IP in IP encapsulation [6]. Optionally foreign agent and home agent may also support GRE and minimal encapsulation. [6] In GPRS, GTP for user plane (which is tunneling protocol as such) is carried over UDP/IP or TCP/IP [3]. TCP/IP is used only if X.25 is carried over GPRS, which most likely will not be widely used commercially. In case of IP over GPRS, GTP carried over UDP/IP.
 

2.1.2. Tunneling End-Point

In Mobile IP, the tunnel from the home agent terminates at the mobile node's care-of address. At the care-of address, the original datagram is removed from the tunnel and delivered to the mobile node. Mobile IP can use two different types of care-of address:  a "foreign agent care-of address" is an address of a foreign agent with which the mobile node is registered, and a "co-located care-of address" is an externally obtained local address which the mobile node has associated with one of its own network interfaces [6]. In case of foreign agent care-of address, the tunnel is terminated at the foreign agent.

In cellular environment, radio resources are scarce. From a radio resource perspective, using of a foreign agent care-of addresses is more efficient. This is because in co-located care-of addresses, the tunnel is terminated at the mobile node [8]. If the tunnel is transported over the radio interface, this means additional IP headed overhead which consumes limited cellular radio capacity. GTP is similar to a foreign agent care-of address, as GTP tunnel is not terminated on mobile station but on SGSN. In GPRS, a tunnel is identified by a tunnel endpoint identifier and a SGSN/GGSN address.
 

2.1.3. Default Router Functionality and Connectivity to External Networks

According to IETF Mobile IP working group, foreign agent is a router on a mobile host's visited network which provides routing services to the mobile node while registered. The foreign agent is capable of detunneling and delivering datagrams to the mobile node that were tunneled by the mobile node's home agent. For datagrams sent by a mobile node, the foreign agent may serve as a default router for registered mobile nodes [6]. In GPRS, as described previously, SGSN is also capable of tunneling and encapsulating user data [3]. SGSN has also some routing functionality, but it is not a default router from application point of view. Practically this means that SGSN is not directly connected to external networks (e.g. Internet). In GPRS, user datagrams are always tunneled to GGSN.

In Mobile IP, the home agent is a router on a mobile node's home network and it does not have to tunnel the packets which are destined to mobile nodes that have direct layer 2 connection. In GPRS, there is no similar "home network" concept from IP routing point of view. GGSN always tunnels IP packets to SGSN.
 

2.1.4. Registration and Tunnel Establishment

In Mobile IP, there is functionality called registration. When the mobile node is away from home, it registers its care-of address with its home agent. This is done through exchange of a Registration Request and Registration Reply messages (which are sent with UDP using well known port number 434). Depending on the method of attachment, the mobile node will register either directly with its home agent, or through a foreign agent which forwards the registration to the home agent. The care-of address can either be determined from a foreign agent's advertisements (a foreign agent care-of address), or by some external assignment mechanism such as DHCP (a co-located care-of address). Mobile IP extends ICMP Router Discovery as its primary mechanism for agent discovery. An agent advertisement is formed by including a mobility agent advertisement extension in an ICMP router advertisement message [6].

According to GPRS specifications, it is possible for a mobile station to be attached to the GPRS network even though it does not have an activated GTP tunnel for data transport. This might be necessary for example when a subscriber wants to be able to receive short messages without using network resources (e.g. IP address). A function similar to Mobile IP registration is built on GPRS specific data connection activation procedure. During this procedure, a GTP tunnel is established between SGSN and GGSN. In Mobile IP (RFC 2002), it is not explicitly described how a tunnels are established. If  IP-in-IP tunneling is used, there can be one tunnel for each IP address pair (e.g. HA and FA).
 

2.2. Convergence between Mobile IP and GPRS Tunneling Protocol

As described in previous section, GPRS core network and Mobile IP are carrying out many similar functions from mobility management point of view. However, the basic Mobile IP specification RFC 2002 as such does not cover all the functionality supported by GPRS core network. 3GPP has carried out a study on combined GSM and Mobile IP mobility handling in UMTS core network. This work aimed on a technical realization where Mobile IP is used to handle mobility in the GPRS and UMTS core networks [8].

Even GTP is not only a mobility management protocol. It also carries out e.g. some session management and tunnel establishment functions. There are some important additions to Mobile IP functionality beyond basic RFC 2002 which are very important when considering GTP replacement by Mobile IP. The following sections describe some of the necessary additions.
 

2.2.1. IP Host Roaming and Accessing External Networks

Network Access Identifier (NAI) is a standardized method for identifying users. It is defined to enhance the interoperability of roaming and tunneling services. Roaming capability in this context may be loosely defined as an ability to use any Internet service provider (ISP). One another need for NAI is secure access to corporate intranets, enabled by tunneling protocols such as PPTP, L2F, L2TP and IPsec in tunnel mode [9]. The NAI has the format similar to an email address which uniquely identifies the user and the user's home network [8]. In GPRS, Access Point Name (APN) has some similar functionality. In GPRS core network, Access Point Name is a reference to the GGSN to be used. Practically APN is used to decide to which IP address GTP tunnel is terminated. In addition, Access Point Name may, in the GGSN, identify the external network. Thus, APN can also be used to determine to which IP address a subsequent tunnel from GGSN (e.g. PPTP, L2F, L2TP or IPsec) is terminated at. Access Point Name is composed of two parts: Network Identifier and Operator Identifier. Network Identifier is a reference to the external network or service. Operator Identifier is a reference to home GPRS network in roaming cases. Both parts of APN are fully qualified domain names according to the DNS naming conventions [3]. NAI and APN have several differences but they support practically the same need: they are tools to give an end user a possibility to describe which external IP network should be accessed. This is important especially in roaming cases. One clear functional difference between NAI and APN is that in addition to a remote network name NAI includes also a user name.
 

2.2.2. PPP Support

RFC 2002 does not specify how Mobile IP is used when mobile nodes connect to their ISP (or intranet) via PPP. RFC 2290 defines a PPP support over Mobile IP [10]. One clear need for PPP is a compatibility with the existing ISP authentication functionality. For example the NAI user id is submitted by the client during PPP authentication process [9]. PPP level authentication is commonly applied in many of the existing ISP remote access servers. By using PPP, one can also carry other packet protocols than IP (e.g. IPX). PPP may also carry IP packets with compressed TCP/IP headers. These PPP options are negotiated during the Network Control Protocol establishment phase [11]. In GPRS, there is similar functionality for transparent PPP tunneling between mobile station and GGSN. PPP can be carried by GTP. Other option is that IP (without PPP) is carried over GTP. If IP over GTP option is applied, some of the PPP specific functionality are lost. A simple request - response authentication based on PPP level functionality can be supported by GPRS also when IP is carried by GTP.
 

2.2.3. Dynamic IP Address Allocation

In RFC 2002 it is assumed that the mobile node includes its (permanent) home address for registration. Also the address of the home agent is included in the registration message and the foreign agent forwards the message to the home agent. The NAI extension described previously has been proposed to handle temporary assignment of home addresses. A mobile node can include a NAI instead of home address in the main part of the Mobile IP Registration Request. Home agent can then allocate IP address to the mobile host using dynamic address allocation principles. The Registration Reply will be sent from the home network to the foreign agent, which extracts the information it needs. This message includes e.g. the IP address allocated for the mobile host [8]. Dynamic address allocation is also an integral part of GPRS. During a GPRS context activation procedure, a GPRS mobile station can indicate whether it requires a static or dynamic address. GGSN can allocate the address internally or use an external AAA server (Authentication, Authorization and Accounting). It is also possible to negotiate the IP address after completion of the context activation procedure. This negotiation is carried out by the mobile station with the external IP network and is likely most feasible in those cases where PPP is carried over GTP.
 

2.2.4. Functions on Cellular Network Operator Level

So far several similar features between GTP and Mobile IP has been identified. In addition to GTP, GPRS core network applies also GSM MAP protocol (Mobile Application Protocol) which is applied between SGSN and HLR (Home Location Register). This protocol supports e.g. authentication between cellular network operator and subscriber, location tracking between cellular network elements and subscriber data transfer. Other cellular networks have similar protocols (e.g. most of the cellular systems defined in America are applying ANSI-41 standard). From backward compatibility point of view, it is not very feasible that these protocols are replaced very quickly. Currently, there no IETF protocol which could immediately replace these protocols.

One requirement for Mobile IP before it can be deployed in future cellular networks is a need to interwork with and operate in co-existence with existing protocols in the cellular networks [12]. To allow Mobile IP to be a mobility solution which supports many different kinds of access networks or technologies, Mobile IP functionality shall be independent of the access network technology [12]. A separation of the authentication procedures is motivated by the fact that radio resources are scarce, and an access network operator may not want to allow Mobile IP signaling until the access network in itself has accepted to provide resources for a mobile node [12].
 

2.2.5. Convergence Status in GPRS/UMTS Release '99 Standards

As a result of these studies there is some convergence between Mobile IP and GTP supported by GPRS release '99 standards. The same applies also to UMTS (Universal Mobile Telecommunications System) release '99. Optionally a foreign agent  functionality can be provided in the GGSN. The interface between the GGSN and foreign agent, including the mapping between the care of IP address and the GTP tunnel is assumed not be standardized as the GGSN and foreign agent are considered to be one integrated node [3]. In principle, one can also implement SGSN and GGSN integrated. However, in the case of inter SGSN handoff, GTP has to be supported anyway. Current Mobile IP specifications cannot be used to ensure lossless handoff [8]. Current Mobile IP specifications don't either support transportation of GPRS subscriber specific information between SGSN's.
 

3. Micro Mobility Issues

IETF does not have very precise definitions for macro mobility and micro mobility. However, as described previously, the basic Mobile IP (RFC 2002) is a specification for macro mobility management. It is also stated that it is less well suited for micro mobility management [6]. Cellular IP on another hand provides tools to solve micro mobility issues in the network layer. Cellular IP supports local mobility, that is, mobility inside an access network. It can interwork with Mobile IP to provide wide area mobility support [7]. However, Cellular IP is not the only way of solving micro mobility issues. Micro mobility can also be handled using link layer mechanisms (i.e. link layer handoff) [6]. IETF has also developed other micro mobility mechanisms such as regional registration. Cellular IP is not yet an RFC but an Internet-draft.

In GSM/GPRS, radio access network consists of base stations and base station controllers. Cellular IP network consists different kinds of nodes. Base station is a Cellular IP node that has a wireless interface. Gateway is a Cellular IP node that is also connected to a regular IP network. Cellular IP Gateway serves also as the mobile host's foreign agent and it relays it's packets both up and downlink [7]. Cellular IP nodes route IP packets inside the Cellular IP Network.

Cellular IP defines also many mobility management concepts which are similar to those applied in existing cellular networks. The following sections describe the micro mobility management in existing IETF documentation and compares it to GPRS. Where needed within the scope of this study, characteristics of WLAN radio access (IEEE 802.11) is compared to the characteristics of GPRS radio access.

3.1. Mobility Management States

Cellular IP introduces a state concept for mobile hosts. A mobile host may either be in active or idle state. A mobile host is in an active state if it is transmitting or receiving IP packets [7]. This state is similar to ready state in GPRS as also in GPRS ready state mobile station may send and receive packets [3]. The other state defined by Cellular IP is called idle and it is applied if a mobile host has not recently transmitted or received IP packets [7]. This state has similarities to standby state in GPRS. In GPRS standby state, data reception and transmission are not possible [3].

In GPRS, ready state is supervised with a timer. A mobile station changes from ready state to standby state when the ready timer expires. The ready timer is reset when data is transferred between SGSN and mobile station [3]. Also in Cellular IP, the mobile station returns to idle state when it has not received or transmitted any data packets for some time [7]. Both in GPRS and in Cellular IP, a state transition timer is not statically defined. These kinds of parameters have typically been a part of cellular network planning. Thus in GPRS, SGSN may change the length of the ready timer by transmitting a new value [3].

Mobility management state functionality is important to minimize the battery consumption in a mobile station. For example a GPRS mobile station in standby state has to monitor only a control channel. In IEEE 802.11 there is no logical channel separation within one carrier. However, IEEE 802.11 standard directly addresses the issue of power saving and defines a mechanism which enables stations to go into sleep mode for long periods of time without losing information. The base station periodically transmits information to power saving stations if they have frames buffered at the base station. If there is a frame buffered at the access point, then the station sends a polling message to the base station to get these frames. As the buffering information is sent periodically, this function requires that stations are synchronized [4]. Some other radio technology specific issues are described in Section 3.3.
 

3.2. Paging Area Concept

Also a concept of paging is introduced in Cellular IP. Mobile hosts that are not actively transmitting or receiving data but want to be reachable for incoming packets maintain paging cache mappings [7]. The IEEE 802.11 standard does not define how roaming should be performed, but defines the basic tools. There is a re-association procedure, where a station which is roaming from one base station to another becomes associated with the new one [4]. Both in GPRS and Cellular IP, the need for paging is related to mobility management states. When a mobile host is in active state, the network must follow its movement from base station to base station to be able to deliver packets without searching for the mobile host. Thus, those hosts which are in active state must notify the network about each handoff [7]. This basic principle applies for both Cellular IP and GPRS.

In Cellular IP, paging area is a set of base stations. Idle mobile hosts crossing cell boundaries within a paging area do not need to transmit control packets to update their position [7]. In GPRS similar functionality is provided by a routing area concept. GPRS mobile station executes mobility management procedures to inform the SGSN when it has entered a new routing area. The mobile station does not inform the SGSN on a change of a cell in the same routing area [3]. Also Cellular IP has a similar functionality: an idle mobile host moving to a new base station transmits a paging update packet only if the new base station is in a new Paging Area [7].

In Cellular IP, wide area mobility occurs when the mobile host moves between Cellular IP networks. The mobile host can identify Cellular IP networks by the Cellular IP network identifier contained in the base stations' beacon signals. The beacon signal also contains the IP address of the gateway. When a mobile host has received this broadcast information, it can send a registration request to the gateway [7]. Mobile host can also send a Mobile IP registration message to its home agent, specifying the gateway's IP address as the care-of address. Alternatively, the gateway can register at the home agent on behalf of the mobile host [7]. In GPRS, a mobile station can initiate a cell re-selection which may change the base station controller (or both base station controller and SGSN). If SGSN is not changed, intra SGSN routing area update procedure is carried out. If SGSN is changed, inter SGSN routing area procedure is carried out. The latter naturally changes the end point of a GTP tunnel which is associated to this mobile station.
 

3.3. Lower Layer Issues

Cellular IP provides some tools to solve micro mobility issues in the network layer. It is not dependent on any specific radio technology. However, Cellular IP does assume that a random access layer 2 protocol covers the air interface. [7] Those layer 2 protocols which are applied over radio interface, typically acknowledge transported data. This is required as radio interface has a relatively high bit error rate. IP as such does not acknowledge received packets and typically it is better to carry out retranmission as soon as possible. For example MAC layer in 802.11 supports both packet acknowledgment and retransmission functionality [4]. Also in GPRS, the radio link control functionality (RLC) provides a reliable link[3].

In IEEE 802.11, one carrier provides a transmission medium which supports 2 Mbps per second in a shared access manner [4]. In many ways, this transmission medium is similar to a segment in a fixed LANs. Instead of collision detection which is used in Ethernet, IEEE 802.11 applies collision avoidance with a positive acknowledge mechanism [4]. From resource reservation point of view, IEEE 802.11 as such has many similarities to fixed LAN technology. Typically no explicit resource allocations are made, but the usage of radio resource is based on shared access principles. If a mobile host does not have anything to send or to receive, it uses radio resources only for control signalling. Also in GPRS, allocation of radio resources is based on the needs for actual packet transfer. A GPRS packet data traffic channel can be allocated for one mobile station at a time. Up to eight packet data traffic channels may be allocated for one mobile station at the same time. GPRS release '97 supports four coding schemes for packet data traffic channels. The data rates of these coding schemes are 9.05 kbps, 13.4 kbps, 15.6 kbps, and 21.4 kbps [13].

In GPRS, there are separate logical channels for transmitting control information and signaling. Some of them are applied in shared access manner but there are also logical channels which are associated only to one mobile station. Random access channel is used to allocate radio resources and it is applied only uplink. Access grant channel is used on opposite direction for radio resource assignment. Also paging has a separate logical channel, which is naturally a downlink only channel [13]. In WLAN there is not this kind of logical channel structure. All control information and signaling is transmitted over the same shared channel.

One important functionality which is carried over radio interface is broadcasting of system information. GPRS applies a separate logical broadcast channels whereas IEEE 802.11 applies beacon frames. All three standards discussed in this paper (Cellular IP, GPRS and IEEE 802.11) define some broadcast information. For example, Cellular IP base stations must periodically transmit beacon signals to allow for mobile hosts to identify an available base station. In each beacon signal, there is a paging area identifier, which enables mobile hosts to notice when they move into a new paging area [7]. Similar functionality is also applied in GPRS for routing areas [3]. Other typical broadcast information in cellular networks include network identification, cell identification and parameters for cell re-selection. In Cellular IP, broadcast information include base station's layer 2 address and gateway's IP address.
 

3.4. Packet Routing

Route Cache is maintained by all Cellular IP nodes and it is used to route packets to mobile hosts. Packets transmitted by the mobile host create and update entries in each node's cache. An entry maps the mobile host's IP address to the neighbor from which the packet arrived to the node. The chain of cached mappings referring to a single mobile host forms a reverse path for downlink packets addressed to the same mobile host. As an active host approaches a new base station, it transmits a route update packet and redirects its packets from the old to the new base station. The route update packet will configure route caches along the way from the new base station to the gateway. Packets transmitted by a mobile host are routed from the base station to the gateway by hop-by-hop shortest path routing, regardless of the destination address [7].

In GSM/GPRS radio access network IP packet is encapsulated over GPRS specific frame structures. Routing of these frames is partially based on topology. One base station is always connected to one base station controller. One base station is always connection to one SGSN. This topology information is used to route the uplink frames. When mobile station is in ready state, SGSN is aware on which cell mobile station is attached. Also base station controller is aware on which cell mobile station is attached. Based on this information it is possible to route the downlink frames to a mobile station in ready state.

In Cellular IP, paging cache is maintained by some Cellular IP nodes and it is used to route packets to mobile hosts. By deploying two caches (paging cache and routing cache), the granularity of location tracking can be different for idle and active mobile hosts. Mobile hosts that are not actively transmitting or receiving but want to be reachable for incoming packets, let their route cache mappings time out but maintain paging cache mappings. IP packets addressed to these mobile hosts will be routed by paging caches. Paging caches have a longer time-out value than route caches and are not necessarily maintained in every node. On the path from the gateway toward the mobile host, the paging packet is broadcast by all nodes it passes. The paging packet is an ordinary IP packet. The set of cells that are reached by the paging packet forms a paging area [7]. In GPRS, if a mobile station is in standby state, SGSN pages it before it transfers data to that mobile station. The paging procedure moves the mobile station to ready state and allows the SGSN to forward downlink data towards the mobile station.

Packet acknowledgment and buffering functionality typically have interactions. If a packet is not acknowledged in certain period of time, it can be retransmitted if it is still stored in a buffer. Buffering can be carried out on different network elements. It is typically preferable that buffering is not carried out very far away from that place where handoff takes place. As described previously, IEEE 802.11 acknowledges on MAC layer. IEEE 802.11 MAC layer is typically implemented in the base station. Thus, one obvious location where buffering can be implemented is base station. How much buffering is to be applied, should be a flexible network configuration parameter. There are certain limits how much delay different higher level protocols and applications can tolerate. For example Voice over IP and streaming technologies require a real time data at a receiver with a strict tolerance. If this tolerance is missed, buffering does not bring any additional benefit. On another hand, some other protocols might tolerate longer packet delays but not any packet loss.

In Cellular IP, when a mobile host switches to a new base station it sends a route update packet to make the chain of cache bindings to point to the new base station. Packets that are traveling on the old path will be delivered to the old base station and will be lost. Although this loss may be small it can potentially degrade TCP throughput. This kind of handoff, when the mobile switches all at once to the new Base Station is called "hard" handoff.  To improve the performance of loss sensitive applications, another type of handoff may be introduced, called "semi soft" handoff. During semi soft handoff a mobile host may be in contact with either of the old and new base stations and receive packets from them. Packets intended to the mobile host are sent to both base stations, so when the mobile host eventually moves to the new location it can continue to receive packets without interruption [7].

3.5. Handoff Decision

Central to the concept of seamless mobility is the process of establishing links at each new connection point. Whenever this process requires the transfer of state information from the old connection point (e.g. a base station) to the new one, a handoff has to occur. There are numerous methods for performing handoff, as numerous as the kinds of state information that has been designed for mobile nodes, as well as the kinds of network entities that maintain the state information [14]. The decision making process of handoff may be centralized or decentralized (i.e. the handoff decision may be made at the mobile station or network) [15]. From decision process point of view one can find at least tree different kind of handoff: As described before, most cellular networks have traditionally been optimized for circuit switched telephony. Especially network controlled and mobile assisted handoff are typical for handling circuit switched bearers (e.g. in NMT, AMPS, GSM, TACS). Mobile IP (RFC 2002) does not specify any explicit control mechanism for a foreign agent to initiate a handoff decision. Mobile IP does not either support transport of any radio interface specific information (e.g. measurement data) on which handoff decision could be based. Cellular IP, on another hand, explicitly states that handoff is initiated by the mobile host [7]. Also in GPRS, a mobile station can autonomously initiate a cell reselection process. The mobile station informs the network when it re-selects another cell [3]. IEEE 802.11 does not specify any explicit control mechanism for any network element to initiate a handoff decision.

There is one potential reasoning why GPRS, Cellular IP and IEEE 802.11 all support mainly mobile controlled handoff. All of them use "on demand" resource reservation principle. This is often reasonable as those resource reservation principles that are applied for circuit switched bearers do not provide optimal radio resource usage for packet switched traffic. Thus, it is not that straightforward to apply similar handoff mechanisms for packet data as for circuit switched bearers. In general, network has to be aware, what kind of bearer or resource reservation is associated to each mobile station. Actually we come back to a previously presented issue in this report. Internet has typically worked by using a best effort traffic model. This has been a very simple and efficient model and several arguments has been stated against any need for a more complicated system [2]. Similar argument may also be applied for wireless Internet access. However, this argument might not be as applicable in cellular environment where radio resources are scarce. In hot spot areas, where higher bit rate WLAN technologies are used, this argument is at least more applicable.
 

5. Conclusions

5.1. Macro Mobility Management

Mobile IPv4 and GTP have overlapping functionality. They are both designed to solve macro mobility issues by tunnelling. Protocol overhead in GPRS is GTP over UDP over IP. IP-in-IP encapsulation used with Mobile IP has overhead of IP over IP. The functionality of IPv6 was not studied in this document, but it provides tools to solve mobility on a single IP protocol level.

GTP is an integral part of first phase GPRS implementations. One possible scenario is that Mobile IP will replace GTP at some point of time. There are several issues beyond RFC 2002, which are needed to fulfil this. Many of these are related on how to access existing external networks. One such GPRS specific mechanism is Access Point Name. Access Point Name is visible to the end user and it has a global significance between operator networks. These all are issues, which make the replacement of functionality more challenging. Backward compatibility with GPRS specific functionalities such as Access Point Name can be one of the trickiest parts of GTP replacement.
 

5.2. Micro Mobility Management

In micro-mobility level, finding similarities is not that straightforward. In GPRS, IP packet is encapsulated over the radio access network. There is no IP level routing functionality within the GPRS radio access network. Thus the “IETF model” which relies heavily on the IP level functionality and extensions has many differences to the GPRS. On the other hand, there is no single “IETF solution” for micro mobility management. One key requirement for micro mobility solutions is that each subsequent registration is not needed to be sent to home agent. One way to solve this problem is hierarchical mobility management. Another key requirement is to smooth the handoff from packet forwarding point of view.

Cellular IP is currently an Internet draft. It can be stated that Cellular IP is more applicable in those environments where hosts are very mobile and the network consists of cells with small radius. The core part of Cellular IP is mobility management states and paging area concept. In cellular network, the main reason for introducing this kind of functionality is reducing the control signaling and limiting the power consumption of the mobile hosts. Cellular IP alone can reduce signaling load but as such it cannot reduce significantly power consumption. It is up to radio access technology how a mobile host can apply a sleep mode functionality. For example in IEEE 802.11 this is carried out by buffering frames in base stations and broadcasting information about these frames periodically in beacon frames. However, from implementation point of view, it can be discussed how well PCMCIA card technology can use this sleep mode functionality. PCMCIA cards consume typically power even if they are only activated from operating system’s point of view.

One key question from micro mobility management’s point of view is that how much functionality really should be carried out in IP protocol level. A typical radio access technology provides layer 2 functionality with error correction, buffering, retransmission and acknowledgement. Handoff functionality can also be built using layer 2 switching functionality. Additional IP level functionality beyond to basic Mobile IP functionality should be justified with clear benefits relative to additional complexity. For example hierarchical mobility management can be used to limit the need for subsequent registrations. Cellular IP can be used to reduce load for control signalling. Some key functionality such as power consumption reduction requires interaction with layer 2 functionality. There are also arguments which are more related to installed base issues than pure technical protocol issues as such. The existing corporate environment relies heavily on layer 2 switching (Ethernet). For example WLAN implementations in the near future will most likely be optimised with this kind of environment. Bringing IP level functionality including management of routing tables at a very low level network element is certainly possible but a significant change to existing situation.
 
 
 

References

  [1] Geoff Huston, ISP Survival Guide. John Wiley & Sons Inc, New York, 1999. ISBN 0-471-31499-4
< Not available online >

[2] Markus Peuhkuri, IP Quality of Service, 5.10.1999.
<http://www.tcm.hut.fi/Opinnot/Tik-110.551/1999/papers/06IPQoS/iwork.html>

[3] 3rd Generation Partnership Project; Technical Specification Group Services and System Aspects;
General Packet Radio Service (GPRS);  Service description; Stage 2 
(3G TS 23.060 version 3.2.1), January 2000
<http://www.3gpp.org/TSG/Dec99_status_list.htm>

[4] Brenner, P., A Technical Tutorial on the IEEE 802.11 Protocol, 18.7.1996, 
< http://www.breezecom.com/pdfs/802.11Tut.pdf >

[5] IETF Mobile IP Working Group, IP Routing for Wireless/Mobile Hosts. Checked at 10th March 2000.
<http://www.ietf.org/html.charters/mobileip-charter.html >

[6] C. Perkins, IP Mobility Support, RFC 2002, October 1996.
<http://www.ietf.org/rfc/rfc2002.txt>

[7] A. Campbel et al, Cellular IP, January 2000.
< http://www.ietf.org/internet-drafts/draft-ietf-mobileip-cellularip-00.txt>

[8] 3rd Generation Partnership Project; Technical Specification Group Services and System Aspects;
Combined GSM and MobileIP Mobility Handling in UMTS IP CN
(3G TR 23.923 version 0.8.0), July 1999
<http://www.3gpp.org/TSG/Dec99_status_list.htm>

[9] B. Aboba et al, The Network Access Identifier, RFC 2486, January 1999.
< http://www.ietf.org/rfc/rfc2486.txt >

[10] J. Solomon et al, Mobile-IPv4 Configuration Option for PPP IPCP, RFC 2290, February 1998.
< http://www.ietf.org/rfc/rfc2290.txt >

[11] W. Simpson et al, The Point-to-Point Protocol (PPP), RFC 1661, July 1994.
< http://www.ietf.org/rfc/rfc1661.txt >

[12] Eva Gustafsson et al, Requirements on Mobile IP from a Cellular Perspective, June 1999.
< http://www.ietf.org/internet-drafts/draft-ietf-mobileip-cellular-requirements-02.txt >

[13] ETSI, Overall description of the GPRS radio interface, Stage 2. GSM 03.64 version 5.2.0.1998.
< Not available online >

[14] Charles E. Perkins, Mobile networking in the Internet. 1998.
 < http://www.baltzer.nl/monet/articlesfree/1998/3-4/mnt071.pdf >

[15] Nishith D. Tripathi, Handoff in Cellular Systems, IEEE Personal Communications. December 1998.
< Not available online >