OPNET Simulation Modeling and Analysis of Enhanced Mobile IP

Taeyeon Park and Arek Dadej

Abstract To facilitate simulation studies of Mobile IP performance and comparative analysis of enhanced Mobile IP handover mechanisms, we have developed a simulation model of Mobile IP using OPNET modeling environment. In this paper, we provide basic design concepts and implementation details of the simulation model, as well as descriptions of the advanced features of IP mobility architectures implemented as part of the model, e.g. buffering and regional registration. Based on the analysis of simulation results obtained using the developed simulation models, a few suggestions are made for the use of Mobile IP and related enhanced mechanisms in selected wireless Internet scenarios.

HA

FA Agent discovery

MN

Data tunneling (fwd)

I. I NTRODUCTION Mobile Internet Protocol (Mobile IP) [11], [13] was developed to provide seamless support for routing of IP datagrams to mobile hosts. The fundamental assumption behind the development of Mobile IP was backward compatibility with the existing Internet infrastructure based on TCP/IP protocol suite originally developed for xed networks. Although Mobile IP was originally aimed at the mobile wireless computing environment built within the realms of the Internet tradition, its principles are also being adopted in the new generation all-IP cellular networks that will replace the traditional mobile telephony networks. To get the maximum benets from deployment of new generation all-IP networks, and to avoid costly design mistakes, it is very important to carefully evaluate Mobile IP and related protocol mechanisms for various usage scenarios and operating conditions that may occur in the next generation mobile Internet. In this paper, we describe implementation of a Mobile IP simulation model. The model covers a number of advanced Mobile IP mechanisms e.g. movement detection [3], [14], buffering [5], [9], [2], [6] and bicasting [11], [7] for handover smoothing, and regional registration [4]. This paper is organized as follows. In Section II, we give a brief overview of Mobile IP and some of its inherent problems. Section III describes design philosophy behind the simulation models of Mobile IP entities. Some basic and enhanced IP mobility mechanisms implemented in the model are explained in the subsequent sections. Sections VI and VII describe the setup of simulation experiments, simulation results and their analysis. In Section VIII, we conclude with the summary of contributions and with comments on the intended future use of the simulation model.
0-7803-7700-1/03/$17.00 (C) 2003 IEEE

Registration / deregistration
Fig. 1. Operation of Mobile IPv4

II. B RIEF OVERVIEW OF M OBILE IP A. Basic Operation The basic sequence of operation for Mobile IPv4 (MIPv4) is as follows: 1) Datagram sent by correspondent node (CN) to mobile node (MN) arrives on home network via standard IP routing. 2) Datagram is intercepted by home agent (HA) and is tunneled to the care-of address. 3) Datagram is detunneled and delivered to the MN. 4) For datagrams sent by the MN, standard IP routing delivers each datagram to its destination. In Fig. 1, the foreign agent (FA) is the MNs default router. The datagram forwarding service in both directions ensures that all datagrams are correctly delivered to their destinations regardless of where the MN currently is. To correctly track the location of the roaming MN, three basic services are needed: agent discovery (to discover the serving FA in a visited network), registration (of the MN with the FA and of the MNs current location with the HA), and movement detection (to detect the need for Mobile IP handover). The interacting entities and the corresponding scope for each of these services are depicted in Fig. 1. B. Known problems and directions for solutions The following issues have been identied within the Mobile IP community as high priority work items that need to be

1017

addressed in order to accelerate widespread wireless Internet deployment and to meet the high service quality standards expected of the next generation wireless Internet. Handover performance fast handover is needed for realtime multimedia trafc and when high frequencies of handovers may be expected [7] Security good protection, e.g. using Authentication, Authorization and Accounting (AAA), from security attacks, is needed in the highly vulnerable wireless environment Integration of Quality of Service (QoS) harmonization with QoS models, such as Differentiated Services (DiffServ) or Multi Protocol Label Switching (MPLS), needs to be achieved in order to guarantee both good mobility support and good quality of service [1] Deployment and migration effective handling of rewall and/or Network Address Translation (NAT) traversal, as well as smooth transition between, or coexistence of, MIPv4 and Mobile IPv6 (MIPv6) In this paper, we focus on the issues in handover performance. III. D ESIGN OF S IMULATION M ODELS This section explains the design philosophy followed in the development of the models, as well as discusses the Mobile IP related functional entities supported in the model. A. Design concepts The general design philosophy followed in the development of our models is modularity and simplicity. Firstly, modularity means ease of extension and exibility in respect to functional changes, thus allowing development of a complete set of models required in the Mobile IP performance studies starting from a very basic set of models. A complete set of models may include implementations of various Mobile IP extension types and a large number of micro/local mobility handling mechanisms that aim at fast handovers and integration with the various avors of QoS control models that may be used in the wireless Internet. Secondly, simplicity helps in focusing on the functional and performance aspects of Mobile IP chosen for the particular study and ensures easy maintenance of the developed model. As an example, the modularity and simplicity of the model mean that by turning on or off one of the IP mobility enhancing mechanisms featured in the model, we can focus on the impact of the selection of this particular enhancement on the Mobile IP signaling load and the overall QoS performance, with minimum or no effort required to change the model to suit the specic Mobile IP networking scenario under investigation. B. Basic entities and modied OPNET library modules Simulation models of Mobile IP entities consist of MN, FA, and HA, together implementing Mobile IP functionality based on [11]. According to the modeling principles used in OPNET, we have constructed each entity with separate Node Model. Each node model consists of several component modules including

processor/queue modules and streams providing connections between modules. Even though most constituents of a node model depend on the type of node, common component modules include udp/tcp transport modules, ip encap/ip rte network modules, and arp/mac modules. Every node should be equipped with more than one physical link, e.g. Ethernet, Point-to-Point, or WLAN (Wireless Local Area Network). The last and most important component module of a node model is the processor module implementing/determining the role of the given node in the network. This is coded into a Process Model. Examples are the MN process model, the FA process model, and the HA process model that are used in the MN, FA, and HA node models respectively. The IPIP process model implements IP-within-IP encapsulation [12] functionality and is used in both FA and HA node models. In addition to developing new node models and process models, we have modied some OPNET library models, such as ip rte/ip encap, arp, and wlan mac modules. In the ip rte module, we need to update the Common IP Routing Table upon receiving a remote interrupt from the MN module to request a change of the Default Gateway. In the arp module, gratuitous and proxy Address Resolution Protocol (ARP) send routines are added and calls to those routines are inserted in a few places in the code. The wlan mac module is modied to contain newly dened Beacon frame formatting and handling, to enable MNs movement among multiple WLAN Access Points (APs) in one OPNET subnet plane. C. Supported functionalities Table I summarizes the functions supported and unsupported in the models. In the table, the abbreviated names for message types are as follows Agent Solicitation (SOL), Agent Advertisement (ADV), Registration Request (RRQ), Registration Reply (RRP), Regional RRQ (R-RRQ), and Regional RRP (R-RRP). IV. BASIC M OBILE IP M ECHANISMS As the basic set of mechanisms for the support of Mobile IP, we implemented several mechanisms including agent discovery and movement detection (ADMD) mechanisms, home registration and deregistration, and IP tunneling mechanism. In the following subsections, we explain each of these mechanisms in detail. A. Agent discovery and movement detection Various kinds of ADMD mechanism are considered in the model. First, Eager Cell Switching (ECS), chosen as a default mechanism in the simulation study, is designed to trigger immediate handover to the new FA as soon as the MN receives an ADV message from that FA. Its maximum ADMD latency equals to the maximum ADV send interval used by the new FA, whereas the mean value approaches half of the maximum. Second, Lazy Cell Switching (LCS) mechanism allows handover to a new FA only after three consecutive packets from the previous FA are lost (the number of lost packets can be varied). As can be easily seen, the maximum

1018

TABLE I S UMMARY OF S UPPORTED F UNCTIONALITY

Features Architectural entities Messages and extensions Care-of address types Encapsulation/tunneling ARP types Replay protection Additional mechanisms

Supported MN, HA, and FA SOL, ADV, RRQ, RRP, R-RRQ, R-RRP, and all extension types FA-COA IP-in-IP (forward tunnel and buffer forward tunnel) ARP, Proxy ARP, and Gratuitous ARP Timestamp-based Enhanced optimal buffering, Regional Registration with hierarchical FA management

Unsupported Collocated-COA Minimal Encapsulation, Generic Routing Encapsulation and IP-in-IP (reverse tunnel) Nonce-based Route optimization, Security and AAA

and mean ADMD latency of the LCS method are four times the max. ADV send interval (under assumption that the radio coverages of the current and the new FAs overlap) and three and a half times the max. ADV send interval, accordingly. Though the ADMD latency for LCS is larger than that of ECS, LCS is still useful in situations where high handover frequency is expected due to high mobility rate and high randomness in mobility pattern. The last method, Active Cell Switching (ACS), is obtained from the basic ECS mechanism by adding layer-2 (L2) triggering of the handover sequence. The MN solicits an ADV message from the new FA upon receiving a L2 trigger notifying that L2 handover has been completed or L2 link to the current FA is about to be broken (the latter depends on the pre-set threshold value of signal strength). The ACS mechanism reduces max. ADMD latency to about an average round trip time (RTT) between the MN and the new FA, which is normally much shorter than the nominal ADV send interval. B. Registration and deregistration Registration mechanism is a core service in the Mobile IP. When MN roams away from its home network and arrives in the radio coverage area of a visited FA, it needs to register its current location with its HA via the FA. While MN is registered with a visited network, it can then issue another registration request to another FA. This is called handover (throughout this paper, the term handover is preferred to the term handoff). The handover results in a change to the active/current mobility binding. In some cases, when simultaneous binding is supported and enabled, the active mobility binding (as opposed to other bindings) can be determined from the context information held by the MN. While MN is away from its home network, the HA acts as a housekeeper to pass all the datagrams destined to the roaming MN. For that purpose, the HA performs gratuitous ARP and proxy ARP on behalf of the MN. Although a basic policy in Mobile IP is to employ soft-state protocol operation, in some cases explicit deactivation of mobility bindings may be required before a timer involved in the soft-state mechanism has expired. For such cases, as well as cases where the MN returns to its home

IPH data

ip_encap
(ENCAP)

IPIP ip_ip(entry) ip_rte Check if registered ?

IPH IPH data

Intercept packets destined to the MN using gratuitous/proxy ARP mechanisms.

Fig. 2.

HA acting as a Tunnel Entry Point

network and wants to remove the mobility bindings, explicit deregistration mechanism may be used. When MN returns home, it should issue gratuitous ARP to let other nodes on the home network know that it has returned. C. Tunneling (Encapsulation and decapsulation) Among several tunneling mechanisms mentioned in [11], we support at present only IP-within-IP [12]; it should be supported in all cases as the default mandatory tunneling mechanism. The operation of the mechanism differs depending on where the mechanism resides (i.e. in HA or in FA). For normal forward tunneling from HA to FA, the IPIP module in the HA node should act as a tunnel entry point as shown in Fig. 2, while the one in the FA node acts as a tunnel exit point as shown in Fig. 3. For the reverse tunneling case, i.e. when the tunnel begins at the FA and ends at the HA, the roles change and are opposite to those in the forward tunneling case. V. A DDITIONAL M OBILITY E NHANCING M ECHANISMS To study performance of different types of Mobile IP handovers, we implemented buffering and bicasting mechanisms as they are widely accepted as handover smoothing techniques.

1019

MN
Protocol==IPIP ?
IPH data

nFA
BUF_CTL_REQ (STO)

oFA

HA/GFA

ip_encap
(DECAP) IPH IPH data

IPIP ip_ip(exit) ip_rte Check if visitor ?

Buffering and direct forwarding of data packets to the MN while in contact with oFA

L2 HANDOVER RRQ Just buffering at oFA BUF_CTL_REQ (FWD)

faVisitorTable

Send to the MN directly using link layer address (hw_addr) through the designated interface.
FA acting as a Tunnel Exit Point

RRQ

Indirect forwarding of buffered data packets from oFA to the MN via nFA
RRP

Fig. 3.

RRP

To reduce registration latency in networks with hierarchical FAs, we have implemented regional registration functionality. In the following subsections, we explain each mechanisms in more detail. A. Optimal buffering sequence Buffering is generally accepted throughout the Mobile IP community as a handover smoothing technique. The term handover smoothing is used here to describe the reduction in packet loss incurred by the handover. To get maximum benet from the use of buffering, we have designed an optimal buffering sequence involving the MN, old/current FA (oFA), new FA (nFA), and HA or gateway FA (HA/GFA), as shown in Fig. 4. The basic principles for controlling when the buffering is done, and for forwarding of the buffered data, are as follows: When a buffering request is made from the MN to the oFA1 via BUF CTL REQ (Store) message, direct forwarding of packets should be carried out as long as the direct communication between the MN and the oFA is possible, in parallel to the buffering at the oFA. This helps eliminate any packet loss that might occur if the overow of buffer pool happens due to its limited size. The BUF CTL REQ (Forward) message is sent just after sending a RRQ, or as an extension of the RRQ message. This ensures minimum delay of the buffered packets delivered to the MN. To avoid packet duplication problem (due to the simultaneous forwarding and buffering at the oFA), last pkt id of the packet that the MN has received last should be sent as a parameter of the BUF CTL REQ (Forward) message. While buffering technique is considered to guarantee zero (or at least minimum) packet loss when designed properly as described in the section above, it can cause unwanted buffering delay for certain classes of user application trafc, such as real-time multimedia trafc.
1 At the time of making the buffering request, the oFA means the current FA servicing the MN.

Direct forwarding via nFA Control message Data packet (Direct)

Data packet (Tunneled) Store into a buffer

An Optimal Buffering Sequence

B. Bicasting with pre-registration Bicasting is another handover smoothing technique well known in the Mobile IP community. Bicasting techniques may be categorized according to the entity that performs bicasting. First and most common case is when the HA (or GFA if regional registration is used) acts as a bicasting entity. In this case the HA/GFA sends the packets destined to the MN both to the oFA and to the nFA. This kind of bicasting can be achieved easily without any additional mechanisms if the basic Mobile IP supports simultaneous mobility binding via S bit of the RRQ message. Another case is when the current FA acts as a bicasting entity. The current FA sends packets both to the MN and to the prospective new FA(s). An example of this kind of bicasting can be found in the Bidirectional Edge Tunnel [7]. Bicasting can reduce greatly the packet loss resulting from large handover latency as buffering technique does. Bicasting is, however, more benecial than buffering in that it also results in the minimum amount of packet delay, comparable with the delay experienced when no handover is made. It should be noted that the minimum packet delay can be achieved only if we make a pre-registration with the (prospective) new FA, using a sort of handover prediction that determines the possible future FA(s) likely to serve the MN. C. Regional registration and hierarchical FA management Regional registration as dened in [4] means a registration local to the visited domain that does not need to go through to the home network unless the MN goes out of the current visited domain. It is believed that such local registration can

1020

reduce Mobile IP related signaling delay, especially when the distance between the visited network and the home network of the MN is large. In Section VII-C, we have compared the overall signaling overhead for the regional and basic (non-regional) registration methods while varying the relative distance between the FA hierarchy and the HA. We have implemented regional registration mechanism along with hierarchical FA management procedures according to [4]. As a part of hierarchical FA management in the FA hierarchy (for example, see Fig. 5), we have modied the ADV message to include the Hierarchical FA Extension, where the hierarchical FA information can be gathered dynamically or congured statically. The original approach dealt with in [4] requires the MN to send explicit request for a regional registration when it receives the ADV with the I bit set. We have developed a modied approach in which FA determines which registration type, home, regional, or global, is appropriate when it receives a normal (meaning global) registration request from the MN. The modied approach does not affect the MN. The FA (or FA group in an FA hierarchy) takes care of all relevant activities by itself. The explicit benet of that approach as compared with the original one is a reduction in size of the ADV message, thus saving in bandwidth consumption over the wireless link. VI. S IMULATION S ETUP Fig. 5 shows the network topology that we have used in our simulations to investigate the basic characteristics of Mobile IP handover mechanisms. In the gure, the R x denotes border routers in each subnetwork, i.e. routers that connect the subnetwork to the Internet. For the home subnetwork, the HA functionality may be incorporated in the border router R h. Similarly, for the foreign subnetwork, the gateway FA functionality that resides in the FA1 may be integrated in the border router R v. In hierarchical terms, the FA1 can act as a gateway FA. Otherwise, it acts as a normal router or normal FA depending on the functionality implemented and the specic needs of the network. The FA hierarchy constructed this way may be used for the purpose of regional registration, or as a at FA topology/structure in other cases. For FAs acting as leaf access routers (FA4 - FA7), it is assumed that the FAs have also been equipped with base station (BS or, in 802.11 terms, AP) functionality. The collocation of FA and BS functionalities in the same node also implies that any number of layer-2 handovers may occur as long as layer-3 IP address (a careof address in Mobile IP sense) has not changed. We have introduced FA-HA path delay 2 to simulate distance between the foreign subnetwork (in the visited domain) and the home subnetwork. To represent that in terms of delay in time domain, we have varied appropriately the delay attribute of the point-to-point link between the border router R v and the Internet Cloud.
2 It reects both geographical distance and abstract distance. While the former is well represented by transmission delay, the latter consists of propagational delay and processing delay at every router along the communication path.

CN

R_c

Internet

R_h

HA

Correspondent Subnetwork R_v FA1 FA2 Visited (Foreign) Domain FA4 MN FA5 FA6

Home Subnetwork

FA3 FA7

Movement Trajectory
Fig. 5. Network Topology Used in the Simulation

Wireless LAN is congured as IEEE 802.11, with 11 Mbps data rate and no RTS/CTS or fragmentation used. Each WLAN radio coverage is set to 250 meters; that ensures non-overlapping radio coverage of separate APs, eventually requiring a sort of hard handover upon crossing the coverage boundaries. Mobility pattern of the MN is characterized by a horizontal linear path with constant ground speed of 30 km/h (the speed has been varied from 1 to 30 km/h when needed to observe the impact of the moving speed on various performance measures). The moving speed (30 km/h) implies that MN moves faster than typical pedestrians but also slower than typical passenger vehicles in a metropolitan area. Consequently, this choice of mobility pattern results in moderate handover rates. The application trafc exchanged between the CN and the MN is congured to represent IP Telephony using Voice-overIP techniques where CN and MN act as clients to each other. The voice trafc exchanged between the MN and CN can start and stop in each direction in a random manner. VII. S IMULATION R ESULTS AND A NALYSIS Using the simulation setup described in the previous section, we have simulated a basic set of handover scenarios and obtained preliminary results. In the simulation, performance measures selected for the purpose of simplied experiments were gathered, namely packet delay, jitter and packet loss during handover. Basic Mobile IP handover operates according to the standard specication dened in [11], whereas handover methods utilizing bicasting and buffering are enabled in addition to the basic Mobile IP handover where required. A. Movement detection performance Fig. 6 shows comparative results of various ADMD methods when basic Mobile IP mechanism without regional registration or buffering is solely used for MN to handover between FAs. As expected from the discussion in Sec. IV-A, LCS method shows the worst performance among the methods compared. ACS is comparable to ECS in performance, giving

1021

900 800 Average handover latency (msec) 700 600 500 400 300 200 100 0

Average Throughput (Kbits/sec)

LCSHOlat ECSHOlat ACSHOlat LCSADMDlat ECSADMDlat ACSADMDlat LCSREGlat ECSREGlat ACSREGlat

10 9.95 9.9 9.85 9.8 9.75 9.7 9.65 9.6 a=1.0, b=0.1 a=1.0, b=0.5 a=1.0, b=1.0 1 5 10 15 20 Moving Speed of MN (Vconst in km/h) 25 30

50

100 150 200 FAHA path delay (msec)

250

300

Fig. 6. Comparison of Average Handover Latency for Various ADMD Methods

Fig. 7.

UDP Application Throughput

a little bit better results at the expense of L2 triggering cost. When considering that overall handover latency is, by denition, the sum of ADMD latency and registration latency, the contribution of ADMD latency only goes below that of the registration latency around 150 msec of FA-HA path delay for the LCS method. For ECS and ACS, it is around 37 and 16 msec respectively. These results imply that the ECS or ACS should be strongly recommended for the scenarios of moderate FA-HA path delay (below 150 msec). B. Mobile IP signaling overhead Fig. 7 and Fig. 8 show the change of application throughput and the corresponding Mobile IP signaling load when the moving speed of the MN at ground level varies from 1 to 30 km/h. For the simulation, no regional registration or buffering other than basic Mobile IP mechanism are used and ECS is applied to decide move detection. In those gures, a and b each mean the send interval of ADV message at layer 3 and Beacon frame at layer 2 in the unit of second. As the moving speed of the MN increases, the overall throughput is going down partly because of the corresponding increase in the signaling load and also partly because of the increase of disconnection time over full communication time due to handover. The higher speed of the MN implies higher handover rate, thus results in the higher disconnected time during which no packet can be exchanged. It also induces larger signaling overhead per unit time period. For xed value of a, move detection method, ECS does not affect overall handover latency. However, the value of b does since layer-3 handover in our simulation setup includes the time for completion of layer-2 handover that can be shortened by shorter value of b. In Fig. 7, as beacon interval b increases throughput decreases more rapidly at higher moving speed of the MN. This tells us that for larger values of b throughput is more affected by communication blackout period (due to frequent handovers). On the other hand, for smaller values of b throughput is more affected by signaling overhead.

350 Average Signaling Load (bits/sec) 300 250 200 150 100 50 0

a=1.0, b=1.0 a=3.0, b=1.0 a=5.0, b=1.0

10 15 20 Moving Speed of MN (Vconst in km/h)

25

30

Fig. 8.

Mobile IP Signaling Load

C. Regional registration performance and FA-HA path delay In Fig. 9 through to Fig. 11, we have shown various performance results for combinations of non-regional/regional registration and basic/buffering cases. Except for the N Ba case, all other methods show comparable results, with the average handover latency 3 around 100 msec when we set 100 msec for a, and 10 msec for b. This proves that our buffering mechanism ensures timely delivery of user trafc as soon as MN begins to register through newly discovered FA. Also, regional registration helps complete local registration in a relatively short time, independent of the value of FA-HA path delay. In Fig. 10, for all cases, throughput decreases as FA-HA path delay increases, since the higher path delay restricts the amount of user trafc along the HA-FA path from CN to MN transferred in a given time interval. Better performance
3 We dene handover latency as difference between time when the rst packet is received from new FA after handover and time when the last packet is received from old FA before handover. If buffering is used, the rst packet through the new FA may be the one buffered at the old FA and then forwarded to the new FA.

1022

700

600 Average handover latency (msec)

NBa NBu RBa RBu

30 NBa NBu RBa RBu

25 Average signaling load (Kbps)

500

20

400 300 200 100

15

10

50

100 150 200 FAHA path delay (msec)

250

300

50

100 150 200 FAHA path delay (msec)

250

300

Fig. 9.

Comparison of Average Handover Latency

Fig. 11.

Comparison of Average MIP Signaling Load

1 0.999 0.998 Normalized throughput 0.997 0.996 0.995 0.994 0.993 0.992 0.991 0.99 0 50 100 150 200 FAHA path delay (msec) 250 300 NBa NBu RBa RBu

Fig. 10.

Comparison of Normalized Throughput

in respect to normalized throughput can be achieved when buffering is used. This is because buffering guarantees zero packet loss regardless of its poor performance in respect to packet delay. Otherwise, regional registration, which reduces handover latency, and thus packet loss during handover, seems to enhance the performance in the case without buffering. While regional registrations can improve performance in situations of higher FA-HA path delay, they also incur about 33 % more average signaling load when three-level FA hierarchy is used as in Fig. 5. The signaling overhead does generally go up as the number of levels in the FA hierarchy increases, even though the increase of overhead is much more dependent on the network topology employed. Some aspects of increased signaling load, such as increased size of the ADV message due to inclusion of the full FA hierarchy information may outweigh the benets of regional registration when bandwidthconstrained wireless links are involved. VIII. C ONCLUSION We have developed and veried a set of Mobile IP simulation models using OPNET network simulation environment.

Preliminary simulation study was undertaken to verify the correctness of the developed simulation models, and to investigate some basic performance characteristics of Mobile IP featuring a few additional handover enhancement mechanisms. Simulation results provide us with some insight as to the characteristics of the considered Mobile IP mechanisms. For example, buffering can be used to guarantee minimum (even zero) packet loss during handover, but it may increase instantaneous packet delay variation undesirable with realtime applications, such as Voice-over-IP. Another example conclusion is that regional registrations should be carefully considered before implemented in the network, since they reduce signaling delay but not signaling load. We are currently researching adaptive mobility control scheme that explores a variety of Mobile IP scenarios and algorithms. This smart IP Mobility control architecture dynamically adapts to the operating conditions and the specic network neighborhood the MN roams into. The developed simulation model is currently undergoing extensions to incorporate additional features necessary to simulate the Adaptive Mobility Control Architecture [10]. ACKNOWLEDGMENT This work was supported by the Commonwealth of Australia through its Cooperative Research Centres Program. We would like to thank anonymous reviewers for their valuable comments in the course of reviewing process. R EFERENCES
[1] H. Chaskar, Editor, Requirements of a QoS Solution for Mobile IP, IETF Internet-Draft, draft-ietf-mobileip-qos-requirements-03.txt, work in progress, July 2002. [2] D. S. Eom, H. S. Lee, M. Sugano, M. Murata and H. Miyahara, Improving TCP handoff performance in Mobile IP based networks, Computer Communications 25(7):635646, May 2002. [3] N. A. Fikouras and C. Gorg, Performance Comparison of Hinted and Advertisement Based Movement Detection Methods for Mobile IP Hand-offs, Computer Networks, 37(1):5562, 2001. [4] E. Gustafsson, A. Jonsson, and C. Perkins, Mobile IPv4 Regional Registration, IETF Internet-Draft, draft-ietf-mobileip-reg-tunnel-06.txt, work in progress, March 2002.

1023

[5] M. Khalil, H. Akhtar, E. Qaddoura, C. E. Perkins, and A. E. Cerpa, Buffer Management for Mobile IP, IETF Internet-Draft, draft-mkhalilmobileip-buffer-00.txt, work in progress, October 1999. [6] G. Krishnamurthi, R. Chalmers, and C. Perkins, Buffer Management for Smooth Handovers in IPv6, IETF Internet-Draft, draft-krishnamurthimobileip-buffer6-01.txt, work in progress, March 2001. [7] MIPv4 Handoffs Design Team, Low Latency Handoffs in Mobile IPv4, IETF Internet-Draft, draft-ietf-mobileip-lowlatency-handoffs-v404.txt, work in progress, June 2002. [8] OPNET Modeler Radio, http://www.opnet.com. [9] S. R. Pandy and S. Jamadagni, Improved Low Latency Handoff in Mobile IPv4, IETF Internet-Draft, draft-shiva-improved-lowlatencyhandoff-v4-01.txt, work in progress, February 2002.

[10] T. Park and A. Dadej, Adaptive Handover between Terrestrial and Satellite Wireless Networks, Proc. CRCSS Conference 2002, Canberra, Australia, pp.46, 1215 February 2002. [11] C. Perkins, Editor, IP Mobility Support for IPv4, IETF, RFC 3344, August 2002. [12] C. Perkins, IP Encapsulation within IP, IETF, RFC 2003, October 1996. [13] J. Solomon, Mobile IP: The Internet Unplugged, Prentice Hall, Englewood Cliffs, 1998. [14] D. Trossen, G. Krishnamurthi, H. Chaskar, J. Kempf, Issues in candidate access router discovery for seamless IP-level handoffs, IETF Internet-Draft, draft-ietf-seamoby-cardiscovery-issues-03.txt, work in progress, June 2002.

1024