A fast handover scheme based on multiple mobile ... - Semantic Scholar

Download PDF
4 downloads 1241 Views 504KB Size Report
Comment
world. To support the newly emerged mobile applications effectively, the mobile network's ... University of Essex, leading the Human-Centred Robotics Group.
202

Int. J. Modelling, Identification and Control, Vol. 10, Nos. 3/4, 2010

A fast handover scheme based on multiple mobile router cooperation for a train-based mobile network Zhiwu Huang* and Yingze Yang School of Information Science and Engineering, Central South University, Changsha Hunan, 410075, China E-mail: [email protected] E-mail: [email protected] *Corresponding author

Huosheng Hu School of Computer Science and Electronic Engineering, University of Essex, Colchester CO4 3SQ, UK E-mail: [email protected]

Kuo-Chi Lin Institute for Simulation and Training, University of Central Florida, Orlando, FL 32826, USA E-mail: [email protected] Abstract: True broadband access for high-speed train passengers has become a new trend in the world. To support the newly emerged mobile applications effectively, the mobile network’s handover latency and packet losses during handover need to be as small as possible. To address this issue, a fast handover scheme based on multiple mobile router cooperation for mobile networks is proposed in this paper. Its basic idea is to take advantage of the features of the fixed line and routing information in the railway passenger transportation system in order to configure the mobile router’s next care-of-address (CoA) in advance, thus reducing the new CoA configuration time. Moreover, two or more mobile routers spatially separated by a certain distance are configured on the railway train that enables different mobile routers to access different subnets during handover and cooperatively receive packets from each other. The overlapped reception of packets from deferent subnets makes the packet loss independent of the handover latency, and reduces packet losses during handover. The mathematical analysis and simulation experiments are presented to show the feasibility and effectiveness of the proposed scheme. Keywords: network mobility; NEMO; fast handover; multihoming; high-speed passenger train. Reference to this paper should be made as follows: Huang, Z., Yang, Y., Hu, H. and Lin, K-C. (2010) ‘A fast handover scheme based on multiple mobile router cooperation for a train-based mobile network’, Int. J. Modelling, Identification and Control, Vol. 10, Nos. 3/4, pp.202–212. Biographical notes: Zhiwu Huang is a Professor in Control Theory and Control Engineering at Central South University. He received his BS in Industrial Automation from Xiangtan University in 1987, his MS in Industrial Automation from Department of Automatic Control, University of Science and Technology Beijing in 1989 and his PhD in Control Theory and Control Engineering from Central South University in 2006. His research interests include fault diagnostic technique and automatic control. Yingze Yang is a PhD student majoring in Control Theory and Engineering in School of Information Science and Engineering, Central South University. He received his BS in Automatisation in 2003 and his MS in Traffic Information Engineering in 2006, both from Central South University, China. His research areas include locomotive network control and network performance optimisation design. Huosheng Hu is a Professor in School of Computer Science and Electronic Engineering at the University of Essex, leading the Human-Centred Robotics Group. His research interests include behaviour-based robotics, human-robot interaction, embedded systems, mechatronics, learning algorithms, pervasive computing, and service robots. He has published over 280 papers in journals, books and conferences in these areas, and received a number of best paper awards. He

Copyright © 2010 Inderscience Enterprises Ltd.

A fast handover scheme based on multiple mobile router cooperation for a train-based mobile network

203

received his MSc from Central South University in China and his PhD from Oxford University in the UK. He is one of founding members of IEEE Robotics and Automation Society Technical Committee on Networked Robots, a Fellow of IET, and a Senior Member of IEEE and ACM. He has been a Chair or Committee Member for many international conferences such as IEEE ICMA, IEEE ROBIO, IEEE IROS, RoboCup Symposiums, and IASTED RA, CA, and CI conferences. He currently serves as one of the Editors-in-Chief for the International Journal of Automation and Computing. Kuo-Chi Lin received his BS and MS in Mechanical Engineering from National Cheng-Kung University, Taiwan, ROC, in 1975 and 1979, respectively. He received his MS and PhD in Aerospace Engineering from University of Michigan, USA, in 1986 and 1990, respectively. Currently, he is an Associate Professor at University of Central Florida, USA. His main research areas include modeling and simulation, multi-agent systems, and collaborative technologies.

1

Introduction

With the emergence of new intelligent systems and robots such as mobile offices, mobile commerce, marine fleet surveillance, and railway passenger/freight transportation, there is a growing need for mobile devices (handheld phones, laptop computers, personal digital assistants, etc.) to maintain access to the internet from anywhere (Cheng et al., 2006; Antonelli, et al., 2006). In many of these mobile devices, their hosts move together as a unit, namely network mobility (NEMO) (Ernst and Lach, 2006). In particular, true broadband access for high-speed train passengers has become a new trend; trains have become a major application platform for mobile networks (Lach et al., 2003). Since train passengers, especially business passengers, need to access the internet, their e-mail and corporate VPNs, it is necessary to manage mobility and provide a continuous internet connection on high-speed trains. However, the existing mobile IPv6 (MIPv6) protocol only supports host mobility, and thus cannot be effectively applied to mobile networks. To address this issue, the mobile network working group of the Internet Engineering Task Force (IETF) has developed the architecture of a mobile network and a NEMO basic support protocol (BNEMO) by extending the MIPv6 protocol (Devarapalli et al., 2005). In general, a mobile network, composed of multiple IP subnets and moving as a single unit on the internet topology, uses mobile routers (MRs) to provide internet connectivity to mobile network nodes (MNNs) via access routers (ARs), as shown Figure 1. MNNs are categorised into three groups: local fixed nodes, local mobile nodes, and visiting mobile nodes. Each MR has at least two interfaces, the egress interface attached to the visited link, and the ingress interface attached to an internal link. MR uses BNEMO to provide uninterrupted connectivity to MNNs. BNEMO builds a bidirectional tunnel between MR and home agent (HA), through which MR and HA communicate with each other. When the MR moves away from its home link, it configures a care-of-address (CoA) in the visited link and registers it with a HA by sending a prefix-scope binding-update (BU) message. Using the received prefix-scope BU messages HA creates a binding cache entry associating MNP with the MR’s CoA. Following the registration, HA intercepts data packets

addressed not only to MR but also to any MNNs that have obtained an address from MNP. It then tunnels the packets to the current location of MR. Similarly, outbound data packets originated from MNNs follow the same route in the reverse direction. Figure 1

NEMO architecture (see online version for colours)

Notes: MNN MR AR HA CN

mobile network node mobile router access router home agent correspondent node

In the train-based BNEMO, an MR can provide internet connectivity to a large number of MNNs. However, as the train moves fast, the MR has to perform handovers frequently. In the IP networks, a handover process generally consists of three sequential operations: link switching, CoA configuration, and binding update. Large handover latency causes the loss of a large number of MNN packets. Therefore the reduction of handover latency is a very important requirement in enabling mobile networks to support multimedia communications. BNEMO, which has large handover delay and high packets loss ratio, cannot meet the requirements for the real-time applications. To address this issue, a fast handover (MMRCFH) scheme based on multiple MR cooperation for the trainbased NEMO is proposed in this paper. By using the fixed

204

Z. Huang et al.

line information and routing information of high-speed passenger train, a subnet link prefix list is maintained in a HA. After it receives the binding update message from MR according to the current CoA, HA obtain the link prefix information of next subnet along the movement direction of train from location identifier of the subnet link prefix list, thus it can generate MR’s next CoA in advance, which can reduce the new CoA (NCoA) configuration time. Moreover, two or more MRs spatially separated by some distance are configured on the train, which enables different MRs to access different subnets during a handover and cooperatively receive packets destined for each other. The overlapped reception of packets from deferent subnets makes the packet loss independent of the handover latency, and tremendous reduces the packet losses during handovers. The remainder of this paper is structured as follows. Section 2 analyses some related handover optimisation scheme for mobile networks. Section 3 describes the handover operation of the proposed MMRCFH scheme in detail. Section 4 and Section 5 present the mathematical analysis results and simulation results respectively. Section 6 concludes the paper and gives a plan for the future work.

2

Handover optimisation for mobile networks

2.1 Fast MIPv6 Handover optimisation schemes for a single mobile node moving on the internet topology have been widely researched in MIPv6-related research. Hierarchical mobile IPv6 (HMIPv6) (Soliman et al., 2005) and fast mobile IPv6 (FMIPv6) (Koodli, 2005) are some of the widely referenced schemes for preventing the performance of a mobile node from being degraded during a handover. The schemes used to reduce handover latency and packet losses include localising signalling domain and predicting mobility in advance to enable a mobile node to complete some part of the handover operation before moving to a new network. These schemes can also be applied to NEMO. Among the above schemes, FMIPv6 is a widely researched and a well-established protocol developed by the IETF for handover optimisation of a mobile node in MIPv6-based networks. In FMIPv6 protocol, packet losses and delays can be reduced by the provision of fast IP connectivity as soon as the new link is established. Using the link-layer (L2) specific mechanism, MN may discover available access points and then request subnet information corresponding to those discovered access points. To reduce the movement detection latency, the FMIPv6 protocol enables an MN to quickly detect that it has moved to a new subnet by providing the new access point and the associated subnet prefix information when the MN is still connected to its current subnet; to reduce the configuration latency, NCoA can be configured in advance through necessary information exchange; to reduce the binding update latency, FMIPv6 specifies a tunnel between the previous CoA (PCoA) and NCoA. As a result, the packet losses and delays of standard MIPV6 are reduced.

There is a strict requirement for the FMIPv6 operations that the neighbouring ARs, i.e., PAR and NAR, have each other’s information, such as the network prefixes, link-layer addresses, and security associations. This is only possible if all ARs belong to the same administrative domain.

2.2 Multihoming-based handover optimisation Mobile networks are more likely to be multihomed than the mobile nodes are. Multihoming refers to a situation where a node can choose between several paths to reach a correspondent. The multiple choices are due to the node having several interfaces to choose from, or the network being connected to the internet by several routers, or by routers with several interfaces. Such a configuration is particularly useful in mobility contexts because it ensures mobile entities remain permanently connected to the internet even in the following situations: loss of connectivity (as a result of moving out of a coverage area or because wireless technologies are more subject to interference), lack of connectivity (a given technology cannot cover all geographic areas) or lack of bandwidth (technologies with high bandwidth are generally not available for mobile users). In addition to the enhanced session continuity, it also allows to choose and balance the traffic between the available connections (Ng et al., 2007). Recently, multi-homing technology has been used by some researchers to enhance the handoff performance of the train-based NEMO. For example, Park et al. (2006) proposed a multihoming-based seamless handover scheme using a multihomed MR with dual antennas in NEMO on railway trains. The two antennas are located at the two far ends of the mobile network. One of the two egress interfaces of MR can continuously receive packets through its antenna while the other is undergoing a handover. Therefore, the proposed scheme has no service disruption or packet losses during handovers. However, the proposed scheme is based on the assumption that all of the MNNs in the mobile network are connected to an MR and only one of these can receive packets during a handover. Since all the traffic of the mobile network has to go through MR, it is difficult to achieve load balancing if it only uses one MR. In addition; the proposed scheme has some overhead in comparison with the basic support. The overhead involves the cost to maintain dual MRs with additional signalling messages. From the above analyses, the existing handover schemes can provide some helpful ideas, but are not for the high-speed passenger train line. They do not make full use of the characteristics such as the fixed passenger railway lines and regular movement of trains. No perfect scheme exists to address the handover latency and packet losses problems caused by frequent handover of the high-speed passenger train. Therefore, there is a need to find an improved scheme that can achieve seamless handover since train-based NEMO has frequent handoffs. Considering the train-based NEMO is moving in a fixed direction for a considerably long time and more likely to be multihomed, it may require two or more MRs to provide internet connectivity to a large number of MNNs. A

A fast handover scheme based on multiple mobile router cooperation for a train-based mobile network multihomed mobile network with widely separated MRs can have simultaneous connections to two or more non-overlapped subnets during a handover. In addition, transport data through different MR-HA tunnels can minimise the packet-loss ratio or handover latency of the real-time communication. Therefore, to meet the needs of train mobile network communications, the multi-homed NEMO having multiple MRs is considered in this work. Both the multihomed NEMO capability of receiving packets from different subnets and the known knowledge of network movement direction are used in designing the MMRCFH scheme.

3

Proposed handover scheme

3.1 System model and assumptions The system model and assumptions of the proposed MMRCFH are described next. Consider a train-based multihomed mobile network: it has many MRs, from which an MNN selects one according to an existing MR selection algorithm. Assume that all MRs are interconnected to each other with good security associations among them. A terminology similar to the one for FMIPv6 is used to describe the scheme: when a MR moves out of a subnet of an AR and enters the subnet of another AR, the former AR is referred as a previous AR (PAR) and the latter AR as a new AR (NAR) of a MR. Similarly, we use the terms of PCoA and NCoA to refer to the MR’s CoAs associated with PAR and NAR, respectively. Figure 2

System model for MMRCFH scheme (see online version for colours)

HA

CN

PAR

MR2

NAR

Distance (d)

MR2 – are considered for simplicity, with MR1 in the front and MR2 in the back of a train-based mobile network, as shown in Figure 2. In this scheme, MNNs in the mobile network can choose to connect to any of MRs, and both MRs can receive data packets during handover, as opposed to the handover scheme proposed in Park et al. (2006) where only one MR can receive data packets. Thus the multiple MRs can achieve load balance. During handover, MR1 leaves the subnet of PAR earlier than MR2, resulting in the mobile network having connectivity with internet through only a single MR2 until MR1 completes its handover process. After that, while MR2 stays in the PAR’s coverage area and MR1 stays in the NARs coverage area simultaneously, the mobile network gets the internet connection through two different subnets simultaneously until MR2 leaves PAR. Note that our scheme can be applied easily to the cases with more than two MRs without any modification. In such a case, each MR should know which MR is its counterpart that can be requested to receive data packets during the handover of the mobile network.

3.2 Handover operation To achieve a low-latency and less-loss handover, the objective of MMRCFH is to obtain NCoA before handover and enable MRs to receive packets on behalf of one another via different subnets during the handover period. For this purpose, it makes full use of the railway passenger transportation characteristics, such as the fixed passenger railway lines and regular movement of trains. As soon as a MR detects that it has moved away from the subnet of PAR, it informs PAR to tunnel its packets to the other MR, which is still located in the subnet of PAR. Similarly, after completing its handover to NAR, the MR helps the other MR to perform the lossless handover to NAR.

A

Internet

MR1 Speed (v)

Prepared work before handover

In order to get the next access network of the train and formulate NCoA in advance, HA maintains a subnet link prefix list, including the IP address of AR in the access network, address prefix length, network link prefix and the location identifier. The location identifier should include the line marker, direction marker and AR sequence number, as shown in Figure 3. Figure 3

Location identifier format of the subnet link prefix list

Line marker

Train-based mobile network

Although the proposed handover scheme can support any number of MRs, in this paper, only two MRs – MR1 and

205

Direction marker

AR sequence number

Here the line marker represents the train line with a two-digit number; the direction marker represents the MR movement direction: 0, the uplink direction and 1, the downlink direction; the AR sequence number is the location of AR computed according to the direction marker. For example, if the marker of Beijing-Guangzhou railway line is 01, then 0110100 represent that the movement direction is

206

Z. Huang et al.

from Beijing to Guangzhou, while the train is in the hundredth access network. When MR leaves the station home network, it registers its home address to HA. According to the MR’s current CoA, HA lookups the current subnet from the subnet link prefix list. Thus it can obtain the link prefix information of next subnet along the train movement direction from the AR sequence number. Then it can formulate MR’s next CoA in advance and send it to PAR. PAR caches it into the CoA table. This operation may be conducted whenever MR completes the HA binding update (HA-BU).

B

Figure 5

Handover operation

(a)

The operation of the proposed handover scheme is depicted in Figure 4, and the data-packet flow paths during handover are shown in Figure 5. Before the handover, both MR1 and MR2 access the internet via PAR. MR1 send an NCoA solicitation (NCoA-Sol) to request PAR to send NCoA. PAR lookups the CoA table, obtains NCoA, and sends it to MR1 by an NCoA acknowledge (NCoA-ACK). In this way, MR1 can obtain NCoA. Figure 4

Data-packet flow paths during handover with MMRCFH scheme (a) MR1 leaves the subnet of PAR (b) MR1 in the new subnet while MR2 still in old subnet (c) MR2 leaves the subnet of PAR (d) both MR1 and MR2 connect to new subnet, next handover waiting (see online version for colours)

(b)

Handover signalling message interactions and datapacket flow paths during handover with MMRCFH scheme (see online version for colours)

(c)

(d)

When MR1 moves out of the subnet of PAR, it executes the following functions: 1

AR-binding update: MR1 sends a binding update message, called AR-binding update (AR-BU), to PAR via MR2, which can easily forward AR-BU to PAR because it is still in the PAR’s subnet. The AR-BU message is similar to a HA-BU message of MR1, except the contents of the home-address destination option and the alternate care-of address fields of the message. In the AR-BU message, the home-address destination option and alternate care-of address option fields contain MR1’s PCoA and MR2’s PCoA, respectively. On receiving the AR-BU message, PAR creates a mapping between MR1’s PCoA and MR2’s PCoA to tunnel the packets addressed to MR1’s PCoA. In other words, when the packets destined for MR1 arrive, PAR encapsulates and forwards these packets to

A fast handover scheme based on multiple mobile router cooperation for a train-based mobile network MR2’s PCoA. MR2 then decapsulates and forwards the packets to MR1 through a wired link. 2

Duplicate address detection and HA-BU: MR1 verifies the uniqueness of NCoA by sending a fast neighbour advertisement (FNA) to NAR and consequently receiving a neighbour advertisement acknowledgement (NAACK). Before sending an NAACK to a MR1, NAR can determine whether NCoA is acceptable. When assigned addressing is used, NAR may assign the proposed NCoA. Such an assigned NCoA must be returned in NAACK. If there is an assigned NCoA returned in NAACK, MR1 must use the assigned address (and not the proposed address in FNA) when attaching it to NAR. Then, MR1 performs the HA registration by sending a HA-BU message, and receiving a binding acknowledgement (HA-BACK) message from HA. Upon the completion of registration, HA tunnels packets to NCoA through NAR.

3

RtAdv relay: After sending the HA-BU message to HA, MR1 relays the RtAdv obtained from NAR to MR2 in an ‘RtAdv relay’ message, so that MR2 can also configure a prospective NCoA for itself, while still remaining connected to PAR.

4

Configuration and cache of MR1’s next CoA: After receive binding update (HA-BU) from MR1, HA send back a binding acknowledgement (HA-BACK) message. According the current CoA of MR, HA lookups the current subnet from the subnet link prefix list to obtain the link prefix information of next subnet along the train movement direction according to the AR sequence number. Then it can formulate MR’s next CoA in advance and send it to NAR (NAR at this time is a PAR relative to the next handover) by a CoA-relay message. NAR then caches it in the candidate CoA table.

5

Operation of MR2: Upon receiving the RtAdv relay message from MR1, MR2 infers that it will need to perform handover to NAR in the near future. MR2 then executes a sequence of functions as performed by MR1. It configures an NCoA from the network prefix of NAR obtained in the RtAdv relay message. It then sends an AR-BU message to NAR via MR1 to verify the uniqueness of NCoA as well as to create a mapping between the MR2’s NCoA and MR1’s NCoA. The creation of this mapping enables MR2 to perform the HA registration using its NCoA before moving to the NAR. Accordingly, MR2 completes the HA registration by sending a HA-BU message to HA and consequently receiving a HA-BACK message.

Following the HA registration, HA forwards the packets destined for MR2 to NCoA. These packets arrive at NAR, which then tunnels them to MR1’s NCoA using the address mapping created by the AR-BU performed by MR2. MR1 decapsulates and forwards the packets to MR2 through a

207

wired link. In this way, having the AR-BU performed prior to the HA registration avoids the possibility of packet losses at NAR. Similarly, by performing the home registration and using NCoA before moving to NAR, MR2 reduces the IP layer handover latency. After receiving packets via MR1, MR2 decides whether to stay in PAR or move to NAR, depending on the availability and strength of the radio signal from NAR in the current location of MR2. When MR2 enters the NAR’s subnet, it can immediately use its already configured and registered NCoA. For this purpose, it has to deactivate its binding at NAR created by AR-BU so that NAR will send packets destined for MR2 to MR2’s NCoA, not to MR1’s NCoA. MR2 deactivates the binding by sending an AR-BU message requesting NAR to create a binding whose lifetime is zero. In other words, to create a binding with a zero lifetime means to delete the binding. From the above explanation and Figure 3, it is clear that the handover operation of MR1 consists of link switching, AR-BU, CoA configuration, and HA registration, whereas the handover of MR2 includes only link switching and ARBU. Although MR1 takes a longer time than MR2 to perform handover, only few packets are lost because of performing AR-BU with PAR.

3.3 Velocity consideration The time duration (Td) during which MR2 remains connected to PAR after MR1 leaves PAR depends on the distance (d) between MR1 and MR2, and the speed (v) of the mobile network, i.e., Td = d/v. For a very-low loss handover, MR2 should remain connected to PAR for duration greater than or equal to the handover latency (Th) of MR1, i.e., Td ≥ Th. For example, suppose a mobile network on a train where d = 100 meters and v = 100 km/hr. The value of Td will thus be 3.6 seconds, which is a sufficient time to allow MR1 to complete binding updates with its HA before MR2 leaves PAR if the mobile network and its home network are located in the same country. Similarly, if the distance between the two MRs is 100 meters and MR can perform handover in one second, our scheme can support a train speed up to 360 km/hr. Conversely, if MR can perform handover in 50 ms and the mobile network moves at a maximum speed of 200 km/hr, the distance between the two MRs can be as small as 2.8 meters. This analysis indicates that for smaller handover latency, our scheme does not necessarily require MRs be separated by a long distance. However, if MRs are separated by a long distance, the scheme can support fairly high velocities and handover latencies. Our scheme is therefore applicable to longer vehicles like the railway trains.

4

Performance analyses

The important metrics in evaluating handover mechanisms are the handover latency and its impact on packet loss. A

208

Z. Huang et al.

mathematical analysis for evaluating and comparing the performance of MMRCFH with BNEMO and FMIPv6 is developed in this section. We chose FMIPv6 for this comparison because it is a widely researched and a well-established protocol developed by IETF for handover optimisation of a mobile node in MIPv6-based networks. Since NEMO is also an extension to the MIPv6 mobility management solution, FMIPv6 is an appropriate candidate protocol to be adopted in NEMO as a handover optimisation solution. For MMRCFH, only MR1 is considered in the evaluation. MR2’s handover latency is very low, consisting of only the link-layer switching time and AR-BU time and therefore not an issue.

4.1 Performance modelling 4.1.1 Handover latencies The handover latency (Th) of MR1 is equal to the sum of the times required for MR1 to carry out the following functions: 1

movement detection and link switching (TL2 + MD)

2

AR-BU via MR2 (TAR – BU)

3

CoA configuration (TCoA)

4

HA-BU (THA – BU).

MMRH TMR = TL 2 + MD + TAR _ BU + TDAD + THA _ BU 1

(1)

To derive the expressions of the remaining component times of equation (1), we use the parameters listed in Table 1.

Bw

+(2Tr +

PHA _ BU + PHA _ BAck Bwl PHA _ BU + PHA _ BAck Bw

+ 2 Lwl ) + 2 Lw )

Parameters definition

(4)

× ( DMR _ HA − 1)

The expression of handover latencies of the basic support protocol is given by NEMO TMR = TL 2 + MD + TCoA + THA _ BU 1

(5)

Here TCoA is the time taken to configure a CoA using RtSolPr and PrRtAdv messages. MR1 verifies the uniqueness of NCoA by sending a neighbour advertisement (NA) to NAR and consequently receiving a NAACK. TCoA = (2Ts +

PRtSol + PRtAdv + 2 Lwl ) + Ts Bwl

P + PNAAck +(2Ts + NA + 2 Lwl ) Bwl

(6)

The handover latency of FMIPv6 is given by equation (7), where TFBU is the time taken to complete fast binding update between PAR and NAR. FMIPv 6 TMR = TL 2 + MD + TFBU + THA _ BU 1

Namely, the handover latency of the proposed scheme is given by

Table 1

THA _ BU = (2Ts +

(7)

Referring to Figure 4, we can derive TFBU as given by equation (8). where x ∈ Y and Y = {FBU, HI, HACK, FBAck}, i.e., a set of the fast binding update, handover indication, handover acknowledgement, and fast binding acknowledgement messages exchanged between PAR and NAR. Note that because the same numbers of signalling messages are involved in executing the predictive and reactive modes of FMIPv6, the handover delays and the signalling overheads in both modes are the same. TFBU = (2Ts +

Bandwidth of wired links

+

∑

PFNA[ FBU ] + PFBAck

(2Ts +

Bwl

+ 2 Lwl )

Px + Lw ) Bw

(8)

Bwl

Bandwidth of wireless links

Lw

Latency of a wired link, i.e., propagation delay + link-layer delay

Lwl

Latency of a wireless link

4.1.2 Packet loss during handover

Ts

Time to configure/process a signalling message

Tr

Routing-table look-up and processing time for a packet in every hop

Px

IP packet length of a signalling message x

DMR_HA

Distance between the MRs and the HA in hops

An analytical model for evaluating the possible number of packet losses during a handover is developed in this subsection. Referring again to Figure 3, during the handover latency given by equation (1), MR1 can not receive packets from the access network only for the first two parts of the latency period, i.e., starting from the instance that MR1 is disconnected from PAR to the instance that PAR receives the AR-BU message sent by MR1 via MR2. Data packets arriving during this time at PAR and destined for MNNs connected through MR1 may be dropped at PAR. The number of packet losses in the case of MMRCFH is thus

TAR _ BU = Ts +

TDAD = 2Ts +

PAR _ BU Bw

+ Lw +

PAR _ BU Bwl

+ Lwl

PFNA + PNAACK + 2 Lwl + Ts Bwl

Y

(2)

(3)

LMMRH = N1λ p (TL 2 + MD + TAR _ BU )

(9)

A fast handover scheme based on multiple mobile router cooperation for a train-based mobile network where N1 is the number of MNNs having active communication sessions through MR1. We next derive the possible number of lost packets in the BNEMO and FMIPv6. In the case of the basic support protocol, MR1 cannot receive packets from its HA during handover until it completes the home-agent binding update. The number of lost packets is therefore given by the following expression. NEMO LNEMO = N1λ pTMR 1

LFMIPv 6 = N1λ p (TL 2+ MD + TFBU )

Handover latencies of different schemes (see online version for colours) 140

Handover latency (ms)

135 130 125 MMRCFH BNEMO FMIPv6

120

115 0

2

4

6

The analyses of the performance of MMRCFH are presented by comparing it with FMIPv6 in terms of the handover latencies and packet losses. To obtain these numerical results, TL2 + MD is set to be 100ms. This value corresponds to an achievable link-layer handover latency of wireless media by employing some optimisations. However, the selection of this value has no significant influence on the IP-layer performance comparisons because it increases the IP-layer handover latencies of all the schemes under consideration by an equal amount. Other parameter values are given in Table 2, which can refer to Banerjee et al. (2003) and Lo et al. (2004). The sizes of the signalling messages are referenced from RFC 3963 (Lach et al., 2003), RFC 4069 (Malamud, 2005), and RFC 2461 (Narten et al., 2007). Some common parameters for evaluation Bw

Bwl

Lw

Lwl

Ts

Tr

100Mbps

11Mbps

0.5ms

2ms

0.8ms

0.001ms

Figure 6 shows the handover latencies of the BNEMO, FMIPv6, and the proposed MMRCFH. It shows that handover latencies of all protocols increase linearly with increasing the distances between the MR and its HA. FMIPv6 has the highest handover latency. While BNEMO does not take any additional time to establish a local forwarding path before performing the HA-BU, FMIPv6 requires a transaction of a number of messages to establish a

8

10

12

14

16

18

20

DMR_HA(hop)

In the MMRCFH and FMIPv6 schemes, MR1 can receive data packets after setting up a local forwarding path from PAR to MR1 (via MR2 in MMRCFH and via NAR in FMIPv6). Figure 7 shows the possible numbers of packets destined for MNNs via MR1 but lost at the PAR during the handover of MR1. This figure shows the lowest number of packets lost in the proposed MMRCFH scheme, which is independent of the distance between MR1 and its HA, because this scheme sets a local forwarding path between PAR and MR1 using a single signalling message. FMIPv6 also establishes a tunnelling path between PAR and MR1, but using a number of signalling messages. The number of packets lost in FMIPv6, although independent of the distance between MR1 and HA and less than that of BNEMO, is more than that of the proposed scheme. Figure 7

Number of packet losses with different schemes versus distance between MR and its HA (see online version for colours) 1400

Number of lost packets

4.2 Discussions

100ms

Figure 6

(11)

Here the assumption is that the MR configures an NCoA while remaining connected to PAR. If the MR moves before configuring NCoA, the time taken to configure NCoA will also be included in the packet loss period.

Table 2

local tunnelling between PAR and NCoA before performing the HA-BU. On the other hand, the MMRCFH scheme has the lowest handover latency, as the NCoA can be obtained in advance. Besides, to establish a local tunnelling, MMRCFH scheme needs only one signalling message from MR1 to PAR via MR2.

(10)

Since the predictive mode of FMIPv6 performs almost all of the handover-related tasks while being connected to the old network, the packet loss may be negligible if the handover prediction is made at an appropriate time. However, making a correct prediction is a difficult task. Therefore, for our comparison we have to take into consideration the packet loss of the reactive mode of FMIPv6. Referring to Figure 4, in the case of FMIPv6, the MR cannot receive its packets until it completes the fast binding update between PAR and NAR. The number of lost packets in FMIPv6 is thus

209

MMRCFH BNEMO FMIPv6

1300

1200

1100

1000 0

2

4

6

8

12

12

14

16

18

20

DMR_HA(hop)

Note: N1 = 100 nodes, λp = 100 pkt/s

In contrast, BNEMO has to perform home-agent binding update, which requires a time proportional to the distance

210

Z. Huang et al.

between MR and its HA in order to enable HA to forward packets to MR via NAR. Until it receives the HA-BU message, HA keeps forwarding the packets to the PAR, which is later dropped from PAR. BNEMO therefore has the highest packet loss that increases with increasing the distance between the MR and its HA. Figure 8 shows that the advantage of having a lower number of packet losses in FMIPv6, compared to the basic support protocol, vanishes as the computation time required by a mobility agent (e.g., MR, AR, or HA) to process or configure a signalling message increasing beyond 5 ms. A message may take longer time to be processed when a mobility agent is overloaded with signalling messages, data packets, or retransmitting functions due to the bad condition of wireless links. The figure shows that although the number of packet losses in the proposed MMRCFH scheme increases slowly with time, it remains lowest for all the values of the processing time. The results clearly indicate that the scheme’s performance remains intact even when the access network is overloaded. Figure 8

Number of packet losses with different schemes versus time required to configure/process a signalling message in a mobility agent (see online version for colours)

2000

Number of lost packets

1800

MMRCFH BNEMO FMIPv6

1600 1400 1200

1000

1 2 3 4 5 6 7 8 Time required to configure/process a signaling message (ms)

Note: N1 = 100, λp = 100pkt/s, DMR_HA = 11 hops

From the above mathematical analysis results one can see the advantages of MMRCFH as follows: 1

Number of signalling minimised: To establish a local tunnel, it requires only one signalling, i.e., an AR-BU message from MR to PAR. It thus improves system scalability.

2

Latency minimised: Since the proposed scheme makes use of the characteristics such as the fixed passenger railway lines and regular movement of the trains, the information of the next access network can be obtained in advance. Therefore, the movement detection latency and NCoA configuration latency can be reduced.

3

Packet losses minimised: It enables a mobile network to receive packets from two or more subnets during the handover period. It therefore reduces the possibilities of packet losses and network service interruption during handover.

The disadvantage of MMRCFH is that] AR must implement a tunnelling mechanism and CoA caching mechanism, while HA must maintain a subnet link prefix list and implement NCoA configuration. But a specific configuration of access network is not required, and the MIPv6-enabled correspondent nodes are left untouched. Besides, it imposes an overhead of an additional encapsulation header attached to data packets over the wireless link between an AR and a MR. However, the encapsulation is mandatory to protect the end-to-end packet integrity. Enduring an overhead is far better than interrupting the network service during the handover period. In summary, the proposed MMRCFH scheme improves the performance of a train-based mobile network during handover by reducing the handover latency and number of packet losses, at the cost of introducing relatively lower amounts of signalling message and packet-delivery overheads in the network. A possible limitation of this study is that it does not consider the dynamics of the link-layer and wireless transmission mechanisms in evaluating the handover latency and packet losses. Nonetheless, modelling the IP-layer behaviour of the proposed scheme and comparing it with other established protocols such as FMIPv6 allow us to draw the conclusion that the scheme is suitable for handover management of train-based mobile networks.

5

Simulation and analysis

To further verify the performance of MMRCFH scheme, we make simulations under the network simulator NS-2 environment (Murray et al., 2008). Some modification and extensions are made on the basis of the FMIPv6 extension module developed by the University of New South Wales and BNEMO extension module developed by Wuhan University, China (Wuhan, 2009). The statistical data of handover latencies and packet loss are obtained, and the simulation results are analysed and compared. Figure 9 shows the network model for the simulation: coverage radius of an AR is 250m, distance between ARs is 400 m, router advertisement interval is one second, the wireless LAN is based on IEEE 802.11 b, and distance between dual antennas is 200 m. Two traffic types are simulated: user datagram protocol (UDP) and transmission control protocol (TCP). For UDP, 512-byte packets were sent repeatedly at a constant rate of 20 packets per second from CN to a MNN residing in the train. Figure 10 shows the packets lost during handover, where the interrupted time is the time of packet loss. Although the handover latencies of MMRCFH is larger than that of the BNEMO, the MMRCFH scheme has the shortest service disruption time of 230 ms, far less than the service interruption time of BNEMO, 2010 ms. Thus the number of lost packets of the MMRCFH scheme, 23 packages, is much less than that of BNEMO, 202 packages, which is about 88.6% reduction. The service interruption time 1040 ms of FMIPv6 is also longer than that of the MMRCFH scheme,

A fast handover scheme based on multiple mobile router cooperation for a train-based mobile network and MMRCFH outperforms FMIPv6 by reducing the packet losses to 77.9%. Figure 9

Network model for simulation (see online version for colours) HA

CN 25ms/10Mbps

25ms/10Mbps

1ms/100Mbps

1ms/100Mbps

v

v

AR

v

AR

MR2

MR1

Movement

Figure 10 Packets lost during handover (see online version for colours) 3600

BNEMO FMIPv6 MMRCFH

3500 3400

UDP ID

3300

TMMRH =230ms

3200

TFMIPv6 =1040ms

3100 3000

TBNEMO =2010ms

2900 34

35

36

37

38

39

Time (s)

Figure 11 Number of packets lost vs. DMR_HA (see online version for colours)

Number of lost packets

Conclusions and future work

A fast handover scheme based on multiple MR cooperation for train-based mobile networks is developed in this paper, which can be used to provide continuous internet connection on the high-speed train for passengers. This scheme takes the advantage of the fixed line features and routing information in the high-speed passengers train to configure the next CoA of MR in advance, which can reduce the NCoA configuration time. Moreover, it exploits the advantage of a multihoming environment of a train-based mobile network; two or more MRs spatially separated by some distance are configured on the train. It thereby makes the packet loss independent of the handover latency by establishing a local tunnel between an AR and a MR by using only one signalling message. The scheme is proved effective through mathematic analysis and simulation experiments. The evaluation results show that the proposed scheme has lower handover latency and packet losses than both BNEMO and FMIPv6. In the future work, we will extend this research by incorporating the dynamics of the link- and physical-layer mechanisms. We will study the buffer-size requirement and additional delays that data packets may face in the AR if some buffering mechanism is used to achieve a lossless handover.

Acknowledgements

300

Many thanks to China CNR Corporation Limited for the financial support.

MMRCFH FMIPv6 BNEMO

250

shown in Figure 11, the simulation results prove the conclusions of performance analysis shown in Section 3.5.2. That is, BNEMO has the highest packet loss that increases as the distance between the MR and its HA grows. While the packet loss of the MMRCFH scheme and FMIPv6 are independent of the distance between MR1 and its HA, and the lowest number of packets are lost in the proposed MMRCFH scheme. It is clear that the simulation results are consistent with the performance analysis in Section 4. The proposed MMRCFH scheme has good feasibility and effectiveness.

6

AR

211

200

References

150

100

50

0 0

1

2

3

4

5

6

7

8

9

10

DMR_HA(hop)

The simulation also tests the impact of DMR_HA on the packet loss by changing the distance between MR and HA, as

Antonelli, G., Arrichiello, F. and Chiaverini, S. (2006) ‘Coordinated control of mobile antennas for ad hoc networks’, International Journal of Modelling, Identification and Control, Vol. 1, No. 1, pp.63–71. Banerjee, N., Wu, W. and Das, S. (2003) ‘Mobility support in wireless internet’, IEEE Wireless Communication, Vol. 10, No. 5, pp.54–61. Cheng, L., Wang, Y.J. and Zhu Q.M. (2006) ‘A communicationbased multirobot rigid formation control system: design and analysis’, International Journal of Modelling, Identification and Control, Vol. 1, No. 1, pp.13–22.

212

Z. Huang et al.

Devarapalli, V., Wakikawa, R. and Petrescu, A. (2005) ‘NEMO basic support protocol’, IETF, RFC 3963, January 2005. Ernst, T. and Lach, H.Y. (2006) ‘Network mobility support terminology draft-ietf-NEMO-terminology-06’, IETF Draft. Koodli, R. (2005) ‘Fast handovers for mobile IPv6’, IETF RFC 4068, July 2005. Lach, H.Y., Olivereau, A. and Motorola, M.M. (2003) ‘Laboratory and field experiments with IPv6 mobile networks in vehicular environments’, Internet-Draft, Draft-lach-nemo-experimentsoverdrive-01.txt, Oct. 2003. Lo, S.C., Lee, G. and Chen, W.T. (2004) ‘Architecture for mobility and QoS support in all-IP wireless networks’, IEEE Journal on Select Areas in Communication, Vol. 22, No. 4, pp.691–705. Malamud, C. (2005) ‘Policy-mandated labels such as adv: in email subject headers considered ineffective at best’, IETF, RFC 4096. Murray, T., Cojocari, M. and Huirong, F. (2008) ‘Measuring the performance of IEEE 802.11p using ns-2 simulator for vehicular networks’, IEEE International Conference on Electro/Information Technology, pp.498–503. Narten, T., Nordmark, E. and Simpson, W. (2007) ‘Neighbor discovery for IP version 6 (IPv6)’, 6IETF, RFC 4861. Ng, C., Ernst, T.E. and Paik. (2007) ‘Analysis of multi-homing in network mobility support’, Internet-Draft, Draft-ietf-nemomultihoming-issues-07, February. Park, H., Kum, D. and Kwon, Y. (2006) ‘IP mobility support with a multi-homed mobile router’, NETWORKING 2006, LNCS 3976, pp.1144–1149. Soliman, H.C., Catelluccia, K. and Malki, E. (2005) ‘Hierarchical Mobile IPv6 Mobility Management’, IETF RFC 4140, August. Wuhan (2009) ‘Simulation for network mobility (NEMO) based on NS2’, avialable at http://www.arcst.whu.edu.cn/center/kongrs/nemo_sim.htm (access on 5 January 2009).