Content uploaded by Samuel Pierre
Author content
All content in this area was uploaded by Samuel Pierre on Sep 12, 2015
Content may be subject to copyright.
972 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 3, MARCH 2008
An Analytical Framework for
Performance Evaluation of IPv6-Based
Mobility Management Protocols
Christian Makaya, Student Member, IEEE, and Samuel Pierre, Senior Member, IEEE
Abstract— Mobility management with provision of seamless
handover is crucial for an efficient support of global roaming
of mobile nodes (MNs) in next-generation wireless networks
(NGWN). Mobile IPv6 (MIPv6) and its extensions were pro-
posed by IETF for IP layer mobility management. However,
performance of IPv6-based mobility management schemes is
highly dependent on traffic characteristics and user mobility
models. Consequently, it is important to assess this performance
in-depth through those two factors. The performance of IPv6-
based mobility management schemes is usually evaluated through
simulations. This paper proposes an analytical framework to
evaluate the performance of IPv6-based mobility management
protocols. This proposal does not aim to advocate which is better
but rather to study the effects of various network parameters on
the performance of these protocols to enlighten decision-making.
The effect of system parameters, such as subnet residence time,
packet arrival rate and wireless link delay, is investigated for per-
formance evaluation with respect to various metrics like signaling
overhead cost, handoff latency and packet loss. Numerical results
show that there is a trade-off between performance metrics and
network parameters.
Index Terms— Analytical modeling, IP mobility protocols,
mobility management, performance evaluation, quality of service,
wireless networks.
I. INTRODUCTION
NEXT-generation wireless networks (NGWN) or fourth
generation wireless networks (4G) are expected to ex-
hibit heterogeneity in terms of wireless access technologies
and services. With NGWN/4G, mobile nodes (MNs) or sub-
scribers will have more demands for seamless roaming across
different wireless networks, support of various services (e.g.,
multimedia applications) and quality of service (QoS) guar-
antees. Conceptually, the NGWN architecture can be viewed
as many overlapping wireless access domains (e.g., UMTS,
CDMA2000, WLAN, WiMAX). However, this heterogeneity
brings new challenges for architecture design, mobility man-
agement, QoS provision and security. Moreover, heterogeneity
in terms of radio access technologies and network protocols
in NGWN requires common interconnection elements. Since
the Internet Protocol (IP) technology enables the support
Manuscript received September 22, 2006; revised November 27, 2006 and
January 22, 2007; accepted March 7, 2007. The associate editor coordinating
the review of this paper and approving it for publication was Y.-B. Lin. This
work was supported in part by the NSERC-Ericsson Industrial Chair.
The authors are with the Mobile Computing and Networking Research
Laboratory (LARIM), Department of Computer Engineering, Ecole Polytech-
nique de Montreal, P.O. Box 6079, Station Centre-ville, Montreal, Quebec,
H3C 3A7, Canada (e-mail: {christian.makaya, samuel.pierre}@polymtl.ca).
Digital Object Identifier 10.1109/TWC.2008.060725.
of applications in a cost effective and scalable way, it is
expected to become the core backbone network of NGWN
[1]. Thus, current trends in communication networks evolu-
tion are directed towards all-IP principles in order to hide
the heterogeneity and achieve convergence of these various
networks.
Mobility management with provision of seamless handoff
is key topic in NGWN/4G. Then, it is crucial to provide
seamless mobility and service continuity in intelligent and
efficient ways. The Internet Engineering Task Force (IETF)
has proposed Mobile IPv6 (MIPv6) [2] as the main protocol
for mobility management at the IP layer. However, MIPv6
has some well known drawbacks such as signaling traffic
overhead, especially when the home agent (HA) or the corre-
spondent node (CN) is located geographically far away from
the mobile node (MN). Message transmission time for binding
update registration will become very high resulting in long
delay (handoff latency) and high packet loss rate thereby
causing user-perceptible deterioration of real-time traffic.
Then, several extensions such as Fast Handovers for MIPv6
(FMIPv6) [3] and Hierarchical MIPv6 (HMIPv6) [4], have
been proposed to enhance MIPv6 performance. In spite of
these extensions, mobility management with QoS provision
in NGWN remains a challenging and complex task. Usually,
performance evaluation of IP-based mobility management
schemes is based on simulation and testbed approaches and
most available work focuses on these aspects [5], [6]. How-
ever, scenarios used for simulations vary greatly, the compar-
ison of IP-based handoff protocols is hardly possible. Few
works are available in the literature which assess IPv6-based
mobility management protocols through analytical models. On
the other hand, they are often based on simple assumptions
and have some drawbacks.
In [7], trade-off relationship between location update cost
and packet tunneling cost is introduced in order to com-
pute total signaling cost and evaluate the efficiency of IP-
based mobility protocols. Work presented in [7] is largely
based on concepts introduced for location management in
personal communication systems (PCS). Analytical models
for handoff latency of IPv6-based mobility protocols are
presented in [8] in order to assess the most appropriate scheme
for functional specification and implementation. Analysis of
signaling bandwidth according to binding update emission
frequency is presented in [9]. However, signaling overhead
generated by packet tunneling is not considered. An analytical
1536-1276/08$25.00 c
2008 IEEE
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
MAKAYA and PIERRE: AN ANALYTICAL FRAMEWORK FOR PERFORMANCE EVALUATION OF IPV6-BASED MOBILITY MANAGEMENT PROTOCOLS 973
model for performance evaluation of HMIPv6 in IP-based
cellular networks was proposed in [10]. This model ignores
periodic binding refresh and binding lifetime period, which
may significantly affect total signaling cost. Moreover, the
packet delivery cost only takes bandwidth consumption into
account for data and ignores the extra signaling consumption
due to control traffic. An analysis of the FMIPv6 signaling
overhead is compared to that of MIPv6 in [11]. However,
packet loss, handoff latency and the impact of user mobility
models were not investigated.
Contrary to previous works, in this paper, we perform
a comprehensive analysis of various IPv6-based mobility
protocols proposed by IETF. We derive signaling overhead
cost, packet delivery cost, binding refresh cost and total
signaling cost generated by an MN during its subnet residence
time for each protocol. Moreover, the required buffer space,
handoff latency and packet loss expressions are derived. The
effect of mobility and traffic parameters on these criteria are
analyzed from numerical results. The remainder of this paper
is organized as follows: the next section offers a brief overview
of IPv6-based mobility management schemes. After that, the
proposed analytical framework is presented. Numerical results
based on this analytical model is then investigated before
concluding remarks drawn in the last section.
II. IP-BASED MOBILITY MANAGEMENT PROTOCOLS
Mobility management enables systems to locate roaming
users in order to deliver data packets, i.e., location manage-
ment and maintain connections with them when moving into
a new subnet, i.e., handover management. Several protocols
have been proposed for these purposes for IP mobility and are
briefly presented in this section.
Definition: Ahandover or handoff is a movement of an
MN between two attachment points, i.e., the process of termi-
nating existing connectivity and obtaining new connectivity.
Handovers in IP-based NGWN may involve changes of the
access point at the link layer and routing changes at the IP
layer.
Efficient mechanisms must ensure seamless handover, i.e.,
with minimal signaling overhead, handoff latency, packet loss,
and handoff failure and service continuity.
Definition: The handoff latency at an MN side is the
time interval during which an MN cannot send or receive any
packets during handoff and it is composed of L2 (link layer)
and L3 (IP layer) handoff latencies. The L3 handoff latency
is the sum of delay due to: movement detection, IP addresses
configuration and binding update procedure.
Definition: The signaling traffic overhead is defined as
the total number of control packets exchanged between an MN
and a mobility agent (e.g., home agent).
A. Mobile IPv6 (MIPv6) [2]
MIPv6 was proposed for mobility management at the IP
layer and allows an MN to remain reachable despite its
movement within the IP environment. Each MN is always
identified by its home address (HoA). While away from its
home network, an MN is also associated with a care-of
address (CoA), which provides information about the MN’s
current location. Discovery of new access router (NAR) is per-
formed through Router Solicitation/Advertisement (RS/RA)
messages exchange. Furthermore, to ensure that a configured
CoA (through stateless or stateful mode [12]) is likely to
be unique on the new link, the Duplicate Address Detection
(DAD) procedure is performed by exchanging Neighbor So-
licitation/Advertisement (NS/NA) messages. After acquiring a
CoA, an MN performs binding update to the home agent (HA)
through binding update (BU) and binding acknowledgment
(BAck) messages exchange. To enable route optimization, BU
procedure is also performed to all active CNs.
However, return routability (RR) procedure must be per-
formed before executing a binding update process at CN
in order to insure that BU message is authentic and does
not originate from a malicious MN. The return routability
procedure is based on home address test, i.e., Home Test
Init (HoTI) and Home Test (HoT) messages exchange, and
care-of address test, i.e., exchange of Care-of Test Init (CoTI)
and Care-of Test (CoT) messages. Although RR procedure
helps to avoid session hijacking, it increases delay of the BU
procedure. Fig. 1(a) represents the sequence of message flow
used in MIPv6 based on stateless address autoconfiguration.
Analysis of MIPv6 shows that it has some well-known
disadvantages such as overhead of signaling traffic, high
packet loss rate and handoff latency, thereby causing user-
perceptible deterioration of real-time traffic. Furthermore, the
scalability problems arise with MIPv6 since it handles MN
local mobility in the same way as global mobility. Simulta-
neous mobility is another problem MIPv6 faces due to route
optimization, which can occur when two communicating MNs
have ongoing session and they both move simultaneously
[13]. These weaknesses have led to the investigation of other
solutions to enhance MIPv6 performance.
B. Fast Handovers for Mobile IPv6 (FMIPv6) [3]
FMIPv6 was proposed to reduce handoff latency and mini-
mize service disruption during handovers pertaining to MIPv6.
The link layer information (L2 trigger) is used either to predict
or rapidly respond to handover events. When an MN detects
its movement toward NAR, by using L2 trigger, it exchanges
Router Solicitation for Proxy (RtSolPr) and Proxy Router
Advertisement (PrRtAdv) messages with the previous access
router (PAR) in order to obtain information about NAR and
to configure a new CoA (NCoA). Then, the MN sends a
Fast Binding Update (FBU) to PAR in order to associate
previous CoA (PCoA) with NCoA. A bi-directional tunnel
between PAR and NAR is established to prevent routing failure
with Handover Initiate (HI) and Handover Acknowledgment
(HAck) message exchanges.
The Fast Binding Acknowledgment (FBAck) message is
used to report status about validation of pre-configured NCoA
and tunnel establishment to MN. Moreover, the PAR estab-
lishes a binding between PCoA and NCoA and tunnels any
packets addressed to PCoA towards NCoA through NAR’s
link. The NAR buffers these forwarded packets until the MN
attaches to NAR’s link. The MN announces its presence on
the new link by sending Router Solicitation (RS) message with
the Fast Neighbor Advertisement (FNA) option to NAR. Then,
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
974 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 3, MARCH 2008
0000011111
00000000001111111111
000000000000000111111111111111
0000011111
00000000001111111111
000000000000000111111111111111
MN NAR HA CN
RA
RS
L2 Handoff
Discovery
Router
NA?
NS
DAD
CoA Configuration
HoTI
HoTI
CoTI
CoT
Handoff
Completed
HoT
HoT
BAck
BU
BU
BAck
RR
(a)
000000000000000111111111111111
000000000000000111111111111111
000000111111
0000011111
0000000000000000000011111111111111111111
0000000000000000000011111111111111111111
RtSolPr
PrRtAdv
FBU
Forward Packets
L2 Handoff
Buffering
FNA
Deliver Packets
BU
L2 Trigger
BAck
CoT
HoTI
CoTI
HoT
HoTI
Handoff
Completed
BU
BAck
CNHA
NARPAR
MN
HoT
HI
HAck
FBAck
FBAck
(b)
Fig. 1. Signaling messages sequence: (a) MIPv6; (b) FMIPv6.
NAR delivers the buffered packets to the MN. The sequence
of messages used in FMIPv6 is illustrated in Fig. 1(b) for
MN-initiated handoff of predictive mode.
A counterpart to predictive mode of FMIPv6 is reactive
mode. This mode refers to the case where the MN does not
receive the FBack on the previous link since either the MN
did not send the FBU or the MN has left the link after sending
the FBU (which itself may be lost), but before receiving
a FBack. In the latter case, since an MN cannot ascertain
whether PAR has successfully processed the FBU, it forwards
a FBU, encapsulated in the FNA, as soon as it attaches to
NAR. If NAR detects that NCoA is in use (address collision)
when processing the FNA, it must discard the inner FBU
packet and send a Router Advertisement (RA) message with
the Neighbor Advertisement Acknowledge (NAACK) option
in which NAR may include an alternate IP address for the MN
to use. Otherwise, NAR forwards FBU to PAR which responds
with FBack. At this time, PAR can start tunneling any packets
addressed to PCoA towards NCoA through NAR’s link. Then,
NAR delivers these packets to the MN.
MN
RA
RS
L2 Handoff
Discovery
Router
NA?
NS
DAD
BAck
BU
Handoff
Completed
LCoA Configuration
MA
P
NAR
(a)
Forward Packets
L2 Trigger
L2 Handoff
RtSolPr
PrRtAdv
FBU
NARMAPPAR
MN@PAR
Buffering
HI
HAck
MN@NAR
FBAck
FBAck
Handoff
LBAck
LBU
Deliver Packets
FNA
Stop Forwarding
Completed
to NAR
(b)
Fig. 2. Signaling messages sequence: (a) HMIPv6; (b) F-HMIPv6.
C. Hierarchical Mobile IPv6 (HMIPv6) [4]
With MIPv6, an MN performs binding update to HA/CNs
regardless of its movements to other subnets. This induces
unnecessary signaling overhead and latency. To address this
problem, HMIPv6 was proposed to handle handoff locally
through a special node called Mobility Anchor Point (MAP).
The MAP, acting as a local HA in the visited network, will
limit the amount of MIPv6 signaling outside its domain and
reduce the location update delay. An MN residing in a MAP’s
domain is configured with two temporary IP addresses: a
regional care-of address (RCoA) on the MAP’s subnet and an
on-link care-of address (LCoA) that corresponds to the current
location of the MN.
As long as an MN moves within MAP’s domain or access
network (AN) it does not need to transmit BU messages to
HA/CNs, but only to MAP when its LCoA changes. Hence,
the movement of an MN within MAP domain is hidden
from HA/CNs. For inter-MAP domain roaming, MIPv6 is
used rather than HMIPv6. When an MN crosses a new
MAP’s domain, moreover from registering with new MAP, BU
messages need to be sent by the MN to its HA/CNs to notify
them of its new virtual location. Fig. 2(a) presents the generic
sequence of message flows used in HMIPv6 with assumption
that an MN has entered into new MAP domain and MIPv6
registration procedure was already completed.
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
MAKAYA and PIERRE: AN ANALYTICAL FRAMEWORK FOR PERFORMANCE EVALUATION OF IPV6-BASED MOBILITY MANAGEMENT PROTOCOLS 975
D. Fast Handover for HMIPv6 (F-HMIPv6) [14]
Combination of HMIPv6 and FMIPv6 motivates the design
of Fast Handover for Hierarchical Mobile IPv6 (F-HMIPv6)
protocol [14] in order to allow more efficient network band-
width usage similarly to HMIPv6. Furthermore, like FMIPv6,
it aims to reduce the handoff latency and packet loss. In
F-HMIPv6, the bi-directional tunnel is established between
MAP and NAR, rather than between PAR and NAR as it
is in FMIPv6. After signaling message exchanges (between
an MN and the MAP) based on FMIPv6 messages, an MN
follows the normal HMIPv6 operations by sending local BU
(LBU) to MAP. When MAP receives LBU with the new LCoA
(NLCoA) from MN, it will stop packets forwarding to NAR
and then clear the established tunnel.
In response to LBU, the MAP sends local BAck (LBAck)
to the MN and the remaining procedure follows the opera-
tions of HMIPv6. In the original F-HMIPv6 proposal, when
handover anticipation cannot be supported, regular operations
of HMIPv6 are used [14]. Hence, HMIPv6 corresponds to
reactive mode of F-HMIPv6. Fig. 2(b) illustrates a sequence
of message used in F-HMIPv6 when an MN moves from
PAR to NAR within MAP’s domain and the MAP already
knows the adequate information on the link-layer address and
network prefix of each AR. This illustration is based on the
assumption that an MN has entered into a new MAP domain
and that MIPv6/HMIPv6 registration procedures were already
completed.
III. ANALYTICAL MODELS
In IPv6-based wireless networks, QoS may be defined
by packet loss, handoff latency and signaling traffic over-
head. Analysis of these metrics is very useful to assess the
performance of mobility management protocols in IP-based
mobile environments. An analytical framework for evaluating
performance of IP mobility protocols is proposed in this
section. The notation used in this paper is given in Table I.
Let χTbe the random variable for the time between L2
trigger generation and link down (i.e., pending L2 handover)
and fT(u, σ)the probability density function for successful
completion of signaling, where σ>0is a success rate pa-
rameter. The probability Psof anticipated handover signaling
success for a particular observed valued tTis expressed as
follows:
Ps=Pr(χT>t
T)=∞
tT
fT(u, σ)du. (1)
Deriving an expression of Psis difficult, as it depends on
the exact form of fT(u, σ), which is usually unknown. For
the sake of simplicity, we assume that χTis exponentially
distributed.
A. User Mobility and Traffic Models
User mobility and traffic models are crucial for efficient
system design and performance evaluation. We consider a
traffic model composed of two levels, a session and packet.
Usually, MN mobility is modeled by the cell residence time
and various types of random variables are used for this purpose
[18]. In NGWN, although the incoming calls or sessions
AR AR
t
t
ij
rs
s
(t )
c
Next callPrevious call
(t )
c
Fig. 3. Timing diagram for subnet boundary crossing.
follow the Poisson process (i.e., inter-arrival time are expo-
nentially distributed), the inter-session arrival times may not
be exponentially distributed [18]. Other distribution models,
like hyper-Erlang, Gamma and Pareto have been proposed to
model various time variables in wireless networks. However,
performance evaluations reported in the literature [18] show
that exponential model can be appropriate for cost analysis.
In fact, exponential model provides an acceptable trade-off
between complexity and accuracy.
Let μcand μdbe the border crossing rate of an MN
out of a subnet (AR) and out of an access network (AN)
or MAP domain, respectively. Furthermore, let μlbe the
border crossing rate for which the MN still stays in the same
AN/MAP domain. When an MN crosses an AN/MAP domain
border, it also crosses an AR border. Then, according to [16],
if we assume that the AN/MAP coverage area is circular with
Msubnets each with size aAR, the border crossing rates are
given by:
μd=μc
√Mand μl=μc−μd=μc
√M−1
√M(2)
where μc=2 v
√πaAR
,vis the average velocity of an MN,
aAR =πR2and Ris the access router radius.
Modeling the probability distribution of the number of
boundary crossing during a call plays a significant role in
cost analysis for wireless cellular networks. This will be the
case again for IP-based wireless networks. Fig. 3 shows the
timing diagram for typical mobile user crossing access router i
(ARi) boundary and moves to ARjduring inter-session time.
trs denotes a residual subnet residence time. In case of inter-
AN/MAP movement, a similar figure for timing diagram for
access network boundary crossing may be drawn by replacing
AR by MAP, tcby tdand trs by the residual access network
residence time (tra).
According to the notation of Table I and the timing diagram
illustration, the subnet crossing probability (Pc) and AN/MAP
domain crossing probability (Pd) during inter-session time
interval are expressed as follows:
Pc=Pr(ts>t
c)=∞
0
Pr(ts>u)fc(u)du
Pd=Pr(ts>t
d)=∞
0
Pr(ts>u)fd(u)du.
(3)
The probability that an MN experiences ksubnets boundary
crossings and maccess network boundary crossings during its
session lifetime corresponds to probability mass function of
Ncand Nd, respectively and expressed as follows [17]:
Pr(Nc=k)=Pk
c(1 −Pc)
Pr(Nd=m)=Pm
d(1 −Pd).(4)
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
976 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 3, MARCH 2008
TAB LE I
NOTATIO N.
tcsubnet (AR’s coverage area) residence time random variable
tdAN/MAP domain residence time random variable
fc(resp. fd)probability density function (PDF) of tc(respectively td)
tsinter-session time between two consecutive sessions with PDF fs
Ncnumber of subnets crossing during intra-AN/MAP handoffs
Ndnumber of AN/MAP domain crossing during inter-AN/MAP handoffs
Cgglobal binding update cost to HA/CNs
Cllocal binding update cost to MAP
Mnumber of subnets in AN/MAP domain
NCN number of CNs having a binding cache entry for an MN
dX,Y average number of hops between nodes Xand Y
CX,Y transmission cost of control packets between nodes Xand Y
PC
Xprocessing cost of control packet at node X
Chc binding update cost at HA and CNs
Crr signaling cost for return routability procedure
tTtime period from the L2 trigger to the starting of link switching
Then, the average number of location binding updates during
an inter-session time interval under subnet crossing (E(Nc))
and AN/MAP domain crossing (E(Nd)) are given by:
E(Nc)=
∞
k=0
kPr(Nc=k)=
∞
k=0
kPk
c(1 −Pc)
E(Nd)=
∞
m=0
mP r(Nd=m)=
∞
m=0
mP m
d(1 −Pd).
(5)
For simplicity and easy derivation of signaling cost, ex-
ponential assumption is made. In other words, we assume
that residence time in a subnet and in AN/MAP domain
follow exponential distribution with parameters μcand μd,
respectively while session arrival process follows a Poisson
distribution with rate λs. Hence, boundary crossing probabil-
ities and average number of location updates during an inter-
session time interval can be easily obtained as follows:
Pc=μc
μc+λs
and Pd=μd
μd+λs
E(Nc)=μc
λs
and E(Nd)=μd
λs
.
(6)
Similarly, we can derive the expression of the average number
of subnets, E(Nl), that an MN crosses but still stay within
AN/MAP domain during an inter-session time interval.
B. Total Signaling Cost
Performance analysis of wireless networks should con-
sider a total signaling cost induced by mobility management
schemes. As for wireless cellular networks, signaling traffic
overhead cost must be evaluated for NGWN or IP-based
mobile environments. In NGWN, there are two kinds of
location update signaling. One occurs from an MN’s subnet
crossing and the other occurs when the binding is about to
expire. To differentiate them, the former refers to binding
update (BU) message and the last one refers to binding refresh
(BR) message. Moreover, delivery of data packets induces
usage of network resources, then generates an additional cost.
Thus, the total signaling cost, CT, could be considered as the
sum of binding update signaling cost, CBU , binding refresh
signaling cost, CBR, and packet delivery cost, CPD:
CT=CBU +CBR +CPD.
Since the signaling cost required for authentication and for L2
handoff are the same for all protocols; then, they are omitted
in our analysis.
C. Binding Update Signaling Cost
Depending on the type of movement and the mobility
management protocol, two kinds of binding updates can be
performed: local and global. For MIPv6 and FMIPv6, global
binding update is performed regardless of movement every
time an MN acquires a new CoA and refers to registration
of CoA to HA and CNs. However, for HMIPv6, global
binding update occurs when an MN moves out of its MAP
domain while local binding update is performed when an
MN changes its current IP address within a MAP domain.
Hence, the average binding update signaling cost for IPv6-
based mobility management schemes during inter-session time
interval depends heavily on the computation of the number of
location binding updates and is given by:
CBU =E(Nl)Cl+E(Nd)Cg.(7)
To perform signaling overhead analysis, a performance fac-
tor called session-to-mobility ratio (SMR), which represents
the relative ratio of session arrival rate to the user mobility
rate, is introduced. The binding update signaling cost becomes:
CBU =1
λsμdCg+μlCl
=1
SMR√MCg+(
√M−1)Cl.
(8)
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
MAKAYA and PIERRE: AN ANALYTICAL FRAMEWORK FOR PERFORMANCE EVALUATION OF IPV6-BASED MOBILITY MANAGEMENT PROTOCOLS 977
The packet transmission cost in IP networks is proportional
to the distance in hops between source and destination nodes.
Furthermore, the transmission cost in a wireless link is gen-
erally larger than the transmission cost in a wired link [7].
Thus, the transmission cost of a control packet between nodes
Xand Ybelonging to the wired part of a network can be
expressed as CX,Y =τdX,Y while CMN,AR =τκ, where
τis the unit transmission cost over wired link and κthe
weighting factor for the wireless link. The global and local
binding update signaling costs for MIPv6 and HMIPv6 are
given by:
Cg
MIPv6=Cl
MIPv6=4CMN,AR +2PC
AR +Chc
Cl
HMIPv6=2(2CMN,AR +PC
AR +CMN,MAP)+PC
MAP
(9)
where Chc is the binding update cost at the HA and at all
active CNs while Crr is the signaling cost due to return
routability procedure. PC
MAP is divided into the mapping
table lookup cost and the routing cost [7], [11].
Let consider one-way transmission cost of HoTI and CoTI
messages during return routability procedure as illustrated in
Fig. 1(a). An MN sends one HoTI message to its HA at
a cost CMN,HA. The HA processes this message at a cost
PC
HA and forwards it to all CNs with NCNCHA,CN as
cost. Each CN processes the received HoTI message before
to respond with HoT message, inducing a processing cost
equal to NCNPC
CN. Then, the cost for home address test
is: 2[CMN,HA +PC
HA +NCNCHA,CN ]+NCNPC
CN.
During the care-of address test, CoTI and CoT messages are
exchanged directly between an MN and CNs. Then, the care-
of address test cost is: 2NCNCMN,CN+NCN PC
CN. We can
then deduce the expression of Crr which is given in Table II.
The link layer information (L2 trigger) is used either to
predict or rapidly respond to handover events in FMIPv6.
Hence, signaling cost of FMIPv6 depends on the probability
that handover anticipation is correct. We assume that if an MN
receives FBAck message from the PAR, then it will definitely
start L3 handover to NAR without exceptions. Hence, if there
is no real handover after L2 trigger, all messages exchanged
from RtSolPr to FBU may be unnecessary. The local binding
update signaling cost for FMIPv6 is expressed as follows:
Cl
FMIPv6=PsSs+(1−Ps)(Sf+Sr)+Chc (10)
where Ssdenotes the signaling cost for a successfully antici-
pated handoff, Sfthe signaling cost for control messages if no
real L3 handoff occurs and Srthe signaling cost for reactive
mode of FMIPv6. Their expressions are given in Table II.
Similar reasoning and assumption as for FMIPv6 allow
computation of signaling cost for F-HMIPv6. The local
binding update signaling cost of F-HMIPv6 is expressed as
follows:
Cl
FHMIPv6=PsSl
s+(1−Ps)Sl
f+Sl
h.(11)
Sl
sand Sl
fhave the same meaning as given above for
FMIPv6 while Sl
his introduced for convenient short form.
Their expressions are given in Table II. FMIPv6 and HMIPv6
can enhance performance of MIPv6 for movement within
AN/MAP domain. However, for inter-AN/MAP movement,
performance of FMIPv6 and HMIPv6 becomes identical to
that of MIPv6. If inter-MAP tunnel is not supported, the same
remarks apply to F-HMIPv6.
D. Binding Refresh Cost
The binding refresh (BR) message is typically used when
the cached binding is in active use but the binding’s lifetime is
close to expiration [2]. Usually, performance analysis available
in the literature did not take into account the periodic binding
refresh and the effect of a binding lifetime period. However,
these parameters may have significant effect on the total
signaling cost. We consider it in our performance analysis and
we propose the binding refresh cost. Let TM,T
Hand TCbe
the binding lifetime period for the MN at MAP, HA and CNs,
respectively. The average rate of sending BR message to MAP
under HMIPv6 while an MN stays in a subnet is 1/(μcTM)
where Xis the integer part of a real number X. By replacing
μcTMwith μdTCand μdTH, respectively, we obtain average
rates of sending BR message to CN and to HA. Hence, the
average binding refresh costs for HMIPv6 and F-HMIPv6 can
be derived as follows:
CHMIPv6
BR =2
1
µcTM
CMN,MAP +
1
µdTH
CMN,HA
+2
1
µdTC
NCNCMN,CN.
(12)
By ignoring the binding refresh cost at MAP, we can obtain
similar expression for MIPv6 and FMIPv6.
E. Packet Delivery Cost
Similarly to [19], we divide handoff latency into three
components: link switching or L2 handoff latency (tL2), IP
connectivity latency (tIP) and location update latency (tU).
IP connectivity latency reflects how quickly an MN can send
IP packets after L2 handoff while location update latency is
the latency of forwarding IP packets to MN’s new IP address.
On the other hand, the time from the starting point of L2
handoff to when an MN first receives IP packets for the first
time after link switching refers to packet reception latency
(tP) or handoff latency. Moreover, we define the following de-
lay components: movement detection delay (tMD), addresses
configuration and DAD procedure delay (tAC), binding update
latency (tBU) and delay from completion of binding update
and reception of first packet at the new IP address (tNR).
Fig. 4 illustrates the timing diagram associated to MIPv6
and shows that there is a delay before an MN begins to receive
packets directly through the NAR.
The packet delivery cost incurs during ongoing session and
is composed of packet transmission and processing costs. The
packet delivery cost could be defined as the linear combination
of packet tunneling cost (Ctun) and packet loss cost (Closs ).
Let αand βbe weighting factors (where α+β=1),
which emphasize tunneling effect and dropping effect; then,
the packet delivery cost is computed as follows:
CPD =αCtun +βCloss.(13)
Let scand sdbe the average size of control packets and data
packets, respectively and η=sd/sc. The cost of transferring
data packet is ηgreater than the cost of transferring control
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
978 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 3, MARCH 2008
TAB L E I I
EXPRESSION OF PARTIAL SIGNALING COSTS.
Sf=3CMN,PAR +2CPAR,NAR +3PC
AR
Ss=4CMN,PAR +3CPAR,NAR +2CMN,NAR +5PC
AR
Sr=2CMN,PAR +2CPAR,NAR +2CMN,NAR +3PC
AR
Sl
f=3CMN,MAP +2(CMAP,NAR +PC
MAP)+PC
AR
Sl
s=4CMN,MAP +3CMAP,NAR +2CMN,N AR +3PC
MAP +2PC
AR
Sl
h=Ps[2(CMN,NAR+CNAR,MAP)+PC
NAR+PC
MAP]+(1−Ps)Cl
HMIPv6
Chc =2(CMN,HA +NCNCMN,CN)+PC
HA +NCNPC
CN +Crr
Crr =2(CMN,HA +NCNCHA,CN +NCNCMN,CN +PC
HA +NCNPC
CN)
latency (t )
Neighbor discovery completes
MN transmission capable
BU received by HA/CNs
Packets begin arriving
Time
t t
NR
Sends binding update (BU)
IP
IP connectivity Link switching
delay (t )
L2
Handover
at the new IP address
start epoch
Packet reception latency (t )
P
Location update latency (t )
t
completed
Movement detection
t
MD AC BU
New link information
U
Fig. 4. Handoff delay timeline of MIPv6.
Packets begin arriving
Time
Neighbor discovery completes
MN transmission capable
Sends binding update (BU)
L2 source
trigger epoch
Forwarding from PAR
to NAR established
directly at the new IP address
Link switching
delay (t )
L2
start epoch information
New link
t
NR
Location update latency (t )
U
Receives buffered packets at NAR
BU
t
BU received by HA/CNs
L2 Handover
Sends RtSolPr
T
t
Obtain NCoA
PN
t
PAR−NAR tunnel
completed
latency (t )
P
Packet reception latency (t )
IP connectivity
P
IP
Fig. 5. Handoff delay timeline of FMIPv6.
packet. Let λpbe the packet arrival rate in unit of packet per
time. There is no forwarding with MIPv6 during handover
(i.e., CMIPv6
tun =0); then, only packet loss cost occurs and is
evaluated as follows:
CMIPv6
loss =λpCf,1
cm (tL2+tIP +tU)(14)
where Cf,1
cm =η(CCN,PAR +CPAR,MN)is the cost of
transferring data packets from CN to MN via PAR when the
handoff fails, tU=tBU +tNR,tBU =tHA+tRR +tCN with
tHA is the delay for performing BU procedure to the HA,
tRR is the delay for return routability procedure and tCN is
the delay for performing BU process to all active CNs.
In HMIPv6, all packets directed to MN will be received
by MAP and after being tunneled to MN’s current address
(LCoA) by using mapping table. Then, the lookup time
of mapping table has an effect on MAP’s processing cost.
Similarly to MIPv6, there is no forwarding with HMIPv6
during handover (i.e., CHMIPv6
tun =0). Hence, packet delivery
cost for intra-AN/MAP roaming can be computed through
(13), where packet loss cost, CHMIPv6
loss , is given by:
CHMIPv6
loss =λpCf,2
cm (tL2+tIP +tL
U)(15)
where tL
Uis the location update latency for intra-AN/MAP
roaming: tL
U=tL
BU +tL
NR with tL
BU the local binding update
latency at MAP while tL
NR is equivalent to tNR for local
roaming and the transferring data packets cost between CN
and MN when the handoff fails is
Cf,2
cm =η(CCN,MAP +CMAP,PAR +CPAR,MN +PC
MAP).
To avoid packet loss, FMIPv6 enables PAR to forward
packets to NAR by using a bi-directional tunnel established
between them and by buffering all forwarded packets. The
timing diagram of predictive mode of FMIPv6 is shown in
Fig. 5 and the packet tunneling cost is given by:
CFMIPv6,p
tun =λpCs,1
cm(tL2+tP
IP +tU)(16)
where Cs,1
cm =η(CCN,PAR +CPAR,NAR +CNAR,MN)is the
cost of transferring data packets from CN to MN by transiting
to PAR and forwarding to NAR via the established tunnel, and
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
MAKAYA and PIERRE: AN ANALYTICAL FRAMEWORK FOR PERFORMANCE EVALUATION OF IPV6-BASED MOBILITY MANAGEMENT PROTOCOLS 979
tP
IP is the IP connectivity latency for predictive mode of fast
handover scheme, tP
IP ≤tIP .
The packet loss due to L2 handoff delay is inevitable
without an efficient buffering mechanism. Moreover, packet
loss in FMIPv6 may be due to wrong temporal and spatial
predictions. Let tPN, the time required to establish a tunnel
between PAR and NAR. Usually, tTis greater than tPN; then,
packets received during handover procedure are forwarded by
PAR to NAR by using the already established tunnel. But,
if MN moves very fast, tTmay be less than tPN. Then,
packets arriving to PAR during the time period tPN −tTmay
be lost, because the tunnel is not yet established. In other
words, for the anticipated signaling to succeed, the following
time constraint must be observed: tPN ≤tT. Hence, packet
loss cost for predictive mode of FMIPv6 can be expressed as
follows:
CFMIPv6,p
loss =λpCf,1
cm max(tPN −tT,0).(17)
Due to wrong spatial prediction of NAR or if FBAck
message was not received on the previous link, the forwarded
packets by PAR may be lost. In this case, the reactive mode
of FMIPv6 is used. Let tR
IP , the IP connectivity latency of
reactive mode. Since the packets forwarding process is not
supported in the reactive mode; then, packet tunneling cost
is equal to zero while packet loss cost for reactive mode of
FMIPv6 can be expressed as follows:
CFMIPv6,r
loss =λpCf,1
cm (tL2+tR
IP +tU).(18)
Hence, the average packet delivery cost of FMIPv6 in terms
of prediction accuracy is given by:
CFMIPv6,a
PD =PsCFMIPv6,p
PD +(1−Ps)CFMIPv6,r
PD .
(19)
With reasoning similar to FMIPv6, evaluation of packet deliv-
ery cost for intra-AN/MAP roaming for F-HMIPv6 is obtained
by replacing tU,tPN,tR
IP ,Cs,1
cm and Cf,1
cm , respectively by
tL
U,tML,tIP ,Cs,2
cm and Cf,2
cm . Where Cs,2
cm is the cost of
transferring data packets from CN to MN by transiting through
the MAP and NAR given by
Cs,2
cm =η(CCN,MAP +CMAP,NAR +CNAR,MN +PC
MAP)
and tML is the time required to establish a tunnel between
MAP and NAR. For inter-AN/MAP roaming, the packet
delivery cost of HMIPv6, FMIPv6 and F-HMIPv6 becomes
the same as for MIPv6.
F. Required Buffer Space
In FMIPv6, NAR buffers packets tunneled from PAR and
forwards them to MN when the latter announces its presence
on the new link. Hence, the required buffer space during
MN’s subnet movement increases in proportion of the packet
arrival rate and according to the number of MNs performing
handover. The buffer space required for FMIPv6 during intra-
AN/MAP handover is proportional to handoff latency and is
computed as follows:
BSl
FMIPv6=λp[Ps(tL2+tP
IP +tU)+(1−Ps)tNR].
(20)
Similarly, buffer space required for F-HMIPv6 is obtained by
replacing tUand tNR by tL
Uand tL
NR in (20), respectively.
Since MIPv6 and HMIPv6 do not use handover anticipation
techniques; then, by setting Ps=0in (20), we obtain a
required buffer space for MIPv6 and HMIPv6.
G. Handoff Latency and Packet Loss
We define the following parameters to compute handoff
latency and packet loss: tL2the L2 handoff latency, tRD
the round-trip time for router discovery procedure, tDAD
the time for DAD process execution, tRR the delay for an
MN to perform return routability procedure and tX,Y one-
way transmission delay of a message of size sbetween
nodes Xand Y. Since the average delay needed for an MN
authentication is the same for all protocols; then, it is omitted.
If one of the endpoints is an MN, tX,Y is computed as follows:
tX,Y (s)= 1−q
1+q
s
Bwl
+Lwl
+(dX,Y −1)
s
Bw
+Lw+q
(21)
where qis the probability of wireless link failure, qthe
average queueing delay at each router in the Internet [20], Bwl
(resp. Bw) the bandwidth of wireless (resp. wired) link and
Lwl (resp. Lw) wireless (resp. wired) link delay. The handoff
latency associated to MIPv6 is given by:
DMIPv6=tL2+tRD +tDAD+tRR +2(tMN,HA +tMN,CN).
(22)
The handoff latency for intra-AN/MAP or localized move-
ment of HMIPv6 is obtained by replacing HA by MAP and
by ignoring tRR and tMN,CN in (22). Let Δns be the time
elapsed from the reception of FBAck on previous link to the
beginning of L2 handoff when there is no good synchroniza-
tion between L2 and L3 handoff mechanisms. Moreover, let
Δlr be the time between last packet reception through previous
link and L2 handoff beginning when FBAck is received on
new link. Note that, Δlr and Δns may be equal to zero
and we use this assumption in performance analysis. For fast
handoff schemes, the handoff latency depends on information
availability, and on which link fast handoff messages are
exchanged. Hence, if information about NAR and impending
handoff are available, and FBAck message is received through
the previous link, handoff latency for localized or micro-
mobility without an efficient buffers management for FMIPv6
and F-HMIPv6 is expressed as follows:
Ol
FMIPv6=Ol
FHMIPv6=Δ
ns +tL2+2tMN,NAR.
(23)
If FBAck message is not received through previous link,
F-HMIPv6 turns to HMIPv6 while for FMIPv6 its reactive
mode is used. Then, handoff latency without efficient buffer
management for FMIPv6 is expressed as follows:
Nl
FMIPv6=Δ
lr +tL2+2tMN,NAR +3tNAR,PAR.
(24)
The average handoff latency for FMIPv6 in terms of prediction
probability is given by:
Dl
FMIPv6=PsOl
FMIPv6+(1−Ps)Nl
FMIPv6.(25)
Similarly, we can obtain the average handoff latency for F-
HMIPv6. The predictive mode of FMIPv6 cannot perform
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
980 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 3, MARCH 2008
TABLE III
SYSTEM PARAMETERS.
Parameters Symbols Val u e s
DAD delay tDAD 500 ms
Router discovery delay tRD 100 ms
L2 handoff delay tL250 ms
Prediction probability Ps0.90
Wireless link failure probability q0.50
Wired link bandwidth Bw100 Mbps
Wireless link bandwidth Bwl 11 Mbps
Wired link delay Lw2ms
Wireless link delay Lwl 10 ms
Number of ARs by AN/MAP M2
Control packet size sc96 bytes
Data packet size sd200 bytes
Packet arrival rate λp10 packets/s
MN average speed v5.6Km/h
Subnet radius R500 m
anticipated IP-handoff for inter-AN [6]; then handoff latency
of FMIPv6 becomes same as for MIPv6. The same remark
applies to HMIPv6 and F-HMIPv6.
With MIPv6 and HMIPv6, packet loss occurs during hand-
off latency or service disruption latency. In fact, the number
of packet loss is proportional to handoff latency. This is also
the case for FMIPv6 and F-HMIPv6 if there is no efficient
buffer management (BM). In fact, for fast handoff schemes
there is no packet loss in theory, unless buffer overflow
happens. Hence, the number of packet lost for each handoff
management scheme is computed as follows:
Pscheme,l
loss =max(BSl
scheme −B, 0) for efficient BM
λpDl
scheme otherwise (26)
where Bis the buffer size of an AR and BSl
scheme is the
buffer space required at an access router for a given scheme
(i.e., MIPv6, HMIPv6, FMIPv6 or F-HMIPv6).
IV. PERFORMANCE EVA L U AT I O N
Parameters and default values used in performance evalu-
ation are given in Table III, except when wireless link delay
and packet arrival rate are considered as variable parameters.
The network topology considered for analysis is illustrated
in Fig. 6, where ER means edge router. For protocols which
do not involve hierarchical mobility management, the MAPs
act as a normal intermediate (edge) router. We assume that
distance (i.e., the number of hops) between different domains
are equals, i.e., c=d=e=f=10and we set a=1,
b=2. The time-to-live (TTL) field in IP packet headers
may be used by an MN to get the number of hops packets
travel. Then, this distance varies within a certain range [7].
All links are supposed to be full-duplex in terms of capacity
and delay. Other parameters used for cost computation are
defined as follows: τ=1,κ=10,α=0.2,β=0.8,σ=2,
PC
AR =8,PC
HA =24,PC
CN =4and PC
MAP =12.
MAP1 MAP2
HA CN
AR1
Movement
MN
b
b
c
d
d
b
f
AR4
AR3AR2
a
b
c
MN’s Home Network
e
CN’s Network
Visited Network
Visited Network
ER
ER
Fig. 6. Network topology used for analysis.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
100
200
300
400
500
600
700
Session−to−Mobility Ratio (SMR)
Signaling Cost
MIPv6
HMIPv6
FMIPv6
F−HMIPv6
Fig. 7. Impact of session-to-mobility ratio on binding update signaling cost.
Most parameters used in this analysis are set to typical values
found in [7], [10], [21].
Fig.7 illustrates the binding update signaling cost during
handoff as a function of SMR for intra-AN/MAP roaming.
When SMR is small, the mobility rate is larger than session
arrival rate; then, an MN changes subnet frequently due
to its mobility, inducing several handoffs and the signaling
overhead increases. However, when the session arrival rate
is larger than mobility rate (i.e., SMR is greater than 1),
binding update is less often performed and signaling overhead
decreases because the frequency of subnet changes decreases.
FMIPv6 and F-HMIPv6 do not effectively reduce signaling
overhead comparatively to MIPv6 and HMIPv6, respectively
due to messages introduced for handoff anticipation. However,
signaling overhead of fast handoff schemes is traded off by
lower handoff latency and packet loss as we will see later.
Fig. 8 represents the effect of binding lifetime period on the
binding refresh cost and shows that the binding refresh cost
decreases as binding lifetime period increases. We assume that
the binding lifetime periods TM,T
Hand TCare equals. We
can see that the binding lifetime period has significant impact
on the average binding refresh cost. Small value of binding
lifetime period leads to larger binding refresh cost; in other
words, significant signaling load throughout the network. On
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
MAKAYA and PIERRE: AN ANALYTICAL FRAMEWORK FOR PERFORMANCE EVALUATION OF IPV6-BASED MOBILITY MANAGEMENT PROTOCOLS 981
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
20
40
60
80
100
120
140
Binding Lifetime Period (hour)
Binding Refresh Cost
MIPv6
HMIPv6
Fig. 8. Impact of binding lifetime period on binding refresh cost.
the other hand, larger value of the binding lifetime period
leads to larger binding cache entry at mobility agents. This
may result in higher memory consumption and higher binding
cache lookup time.
The result shows that the binding refresh cost remains
constant when the binding lifetime period is between 0.4and
0.7hour and as well as when it is greater than 0.8hour. For
the former case, the result indicates that during [0.4,0.7] time
period there is the same number of binding refresh messages.
This is due to the fact that an MN moves to an adjacent subnet
before the new binding refresh message occurs. While for the
latter case, the average subnet residence time of an MN is
shorter than the binding lifetime period (i.e., TM≥0.8hour).
Hence, no binding refresh message occurs and the binding
refresh cost is equal to zero. On the other hand, due to binding
cache and lookup table maintained at the MAP, there is an
extra cost for binding refresh process at the MAP for HMIPv6.
Thus, binding refresh cost of HMIPv6 is slightly greater than
for MIPv6.
The packet delivery cost is depicted in Fig. 9 as a function of
packet arrival rate (λp). We observe that, packet delivery cost
increases proportionally with λpfor all schemes. Fast handoff
schemes (i.e., F-HMIPv6 and FMIPv6) outperform MIPv6 and
HMIPv6, and they are more efficient when λpincreases. This
means that FMIPv6 and F-HMIPv6 are better suited for real-
time applications where periodic packets are sent at high rates.
The packet delivery cost depends on handoff latency, while
packet loss is proportional to handoff latency. Then, a similar
analysis may be performed for packet loss when comparing
to packet arrival rate as in Fig. 9. Hence, packet loss will be
lesser for fast handoff schemes than for MIPv6 and HMIPv6.
For varying prediction probability, Ps,Fig.10showsthe
behavior of packet delivery cost. The packet delivery cost
decreases when the accuracy of Psincreases for fast handoff
schemes. Due to additional packet processing at MAP for
F-HMIPv6, there is an extra cost for packet delivery with
inaccuracy prediction. In fact, in this case, F-HMIPv6 turns to
HMIPv6, as we can see when Ps=0. HMIPv6 and MIPv6
12345678910
0
10
20
30
40
50
60
70
Packet Arrival Rate (packet/s)
Packet Delivery Cost
MIPv6
HMIPv6
FMIPv6
F−HMIPv6
Fig. 9. Packet delivery cost as a function of packet arrival rate.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
10
20
30
40
50
60
70
Prediction Probability (Ps)
Packet Delivery Cost
MIPv6
HMIPv6
FMIPv6
F−HMIPv6
Fig. 10. Packet delivery cost as a function of prediction probability.
are not affected by the prediction probability. For high values
of Ps, F-HMIPv6 performs better than FMIPv6. Since there
is a relation between handoff latency and packet delivery cost,
a similar behavior will be observed when comparing handoff
latency with prediction probability. Hence, an effective pre-
diction mechanism is required to allow better performance for
F-HMIPv6.
To alleviate packet losses, fast handover schemes should
support packet buffering and forwarding during handoff ex-
ecution. Since fast handover schemes start packet buffering
and forwarding earlier; then, they require more buffer space
than MIPv6 and HMIPv6 as we can see in Fig. 11. On the
other hand, buffering time may affect real-time applications,
for example if some packets are stored in a buffer for a
longer period of time than acceptable end-to-end delay, they
may become useless. Hence, it is crucial to manage buffers
efficiently in order to minimize overhead and to provide better
QoS to delay sensitive applications.
In Fig. 12, we can see that the handover latency increases
proportionally with the wireless link delay. We observe that
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
982 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 3, MARCH 2008
12345678910
0
100
200
300
400
500
600
Packet Arrival Rate (packet/s)
Buffer Space (bytes)
MIPv6
HMIPv6
FMIPv6
F−HMIPv6
Fig. 11. Required buffer space as a function of packet arrival rate.
010 20 30 40 50 60 70 80
0
0.5
1
1.5
Wireless Link Delay (ms)
Handoff Latency (s)
MIPv6
HMIPv6
FMIPv6
F−HMIPv6
Fig. 12. Impact of wireless link delay on handoff latency.
MIPv6 and HMIPv6 have worst results among all protocols
followed by FMIPv6 while F-HMIPv6 performs better than
all other schemes. For MIPv6 and HMIPv6, the DAD process
counts for a large portion of handoff delay. Therefore, it is im-
portant to decrease the DAD delay in order to decrease handoff
latency. The optimistic DAD (oDAD) [22] has recently been
proposed to allow minimization of address configuration delay
by eliminating the DAD completion time.
V. C ONCLUSION
Mobility management is a key issue in next-generation
or 4G wireless networks (NGWN/4G). Several IPv6-based
mobility schemes have been proposed in the literature and
by IETF. However, they are not able to guarantee seamless
roaming and service continuity for critical applications like
real-time applications. Moreover, performance evaluation of
these schemes is usually based on simulation approaches.
This paper proposes a comprehensive analytical model for
IPv6-based mobility protocols (i.e., MIPv6, HMIPv6, FMIPv6
and F-HMIPv6) in order to provide depth analysis of the
overall performance of these protocols. Several performance
metrics such as signaling overhead cost, packet delivery cost,
handoff latency and packet loss are analyzed according to user
mobility and traffic models. Our goal was not to decide which
scheme is always better, but to study the effect of various
parameters related to mobility and traffic on the performance
of these schemes in order to facilitate decision-making for
wireless network design.
The numerical results show the potential pros and cons
of most promising IPv6-based mobility schemes proposed by
IETF. They reveal that F-HMIPv6 enables improvement in
terms of handoff latency and packet loss rather than other
protocols (i.e., MIPv6, HMIPv6 and FMIPv6). However, this
performance is off-set by its signaling traffic overhead and the
buffer space required when compared to HMIPv6. Moreover, it
is very difficult to forecast which IPv6-based mobility protocol
will dominate in NGWN/4G. In fact, selection of a mobility
management scheme is not based solely on performance cri-
teria, but on cost and respective profits as well. Thus, until an
ideal mobility management protocol is designed and deployed,
mobile users still require a practical solution. This could be
achieved by a certain tradeoff of the above requirements.
ACKNOWLEDGMENT
The authors would like to thank the anonymous reviewers
for their helpful feedback, suggestions and comments to
improve the presentation of this paper.
REFERENCES
[1] I. F. Akyildiz, S. Mohanty, and J. Xie, “A ubiquitous mobile communi-
cation architecture for next-generation heterogeneous wireless systems,”
IEEE Commun. Mag., vol. 43, no. 6, pp. 29-36, June 2005.
[2] D. B. Johnson, C. E. Perkins, and J. Arkko, “Mobility support in IPv6,”
IEFT RFC 3775, June 2004.
[3] G. Koodli, “Fast handovers for mobile IPv6,” IETF RFC 4068, July
2005.
[4] H. Soliman, C. Castelluccia, K. El-Malki, and L. Bellier, “Hierarchical
mobile IPv6 mobility management (HMIPv6),” IETF RFC 4140, Aug.
2005.
[5] X. P´
erez-Costa, M. Torrent-Moreno, and H. Hartenstein, “A performance
comparison of mobile IPv6, hierarchical mobile IPv6, fast handovers
for mobile IPv6 and their combination,” ACM Mobile Computing and
Commun. Rev., vol. 7, no. 4, pp. 5-19, Oct. 2003.
[6] Y. Gwon, J. Kempf, and A. Yegin, “Scalability and robustness analysis
of mobile IPv6, fast mobile IPv6, hierarchical mobile IPv6, and hybrid
IPv6 mobility protocols using a large-scale simulation,” in Proc. IEEE
Int. Conf. on Commun. (ICC’04), vol. 7, June 2004, pp. 4087-4091.
[7] J. Xie and I. F. Akyildiz, “A novel distributed dynamic location
management scheme for minimizing signaling costs in mobile IP,” IEEE
Trans. Mobile Computing, vol. 1, no. 3, pp. 163-175, July/Sept. 2002.
[8] X. P´
erez-Costa, R. Schmitz, H. Hartenstein, and M. Leibsch, “A MIPv6,
FMIPv6 and HMIPv6 handover latency study: analytical approach,” in
Proc. IST Mobile and Wireless Commun. Summit, June 2002, pp. 100-
105.
[9] C. Castelluccia, “HMIPv6: a hierarchical mobile IPv6 proposal,” ACM
Mobile Computing and Commun. Rev., vol. 4, no. 1, pp. 48-59, Jan.
2000.
[10] S. Pack and Y. Choi, “Performance analysis of fast handover in mobile
IPv6 networks,” in Proc. IFIP Pers. Wireless Commun., LNCS, vol.
2775, Sept. 2003, pp. 679-691.
[11] S. Pack and Y. Choi, “A study on performance of hierarchical mobile
IPv6 in IP-based cellular networks,” IEICE Trans. Commun., vol. E87-
B, no. 3, pp. 462-469, March 2004.
[12] S. Thomson and T. Narten, “IPv6 stateless address autoconfiguration,”
IETF RFC 2462, Dec. 1998.
[13] K. D. Wong, A. Dutta, H. Schulzrinne, and K. Young, “Simultaneous
mobility: analytical framework, theorems and solutions,” Wireless Com-
mun. and Mobile Computing (Wiley), to appear.
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
MAKAYA and PIERRE: AN ANALYTICAL FRAMEWORK FOR PERFORMANCE EVALUATION OF IPV6-BASED MOBILITY MANAGEMENT PROTOCOLS 983
[14] H. Y. Jung, E. A. Kim, J. W. Yi, and H. H. Lee, “A scheme for supporting
fast handover in hierarchical mobile IPv6 networks,” ETRI Journal,vol.
27, no. 6, pp. 798-801, Dec. 2005.
[15] W. Wang and I. F. Akyildiz, “Intersystem location update and paging
schemes for multitier wireless networks,” in Proc. ACM MOBICOM,
Aug. 2000, pp. 99-109.
[16] F. V. Baumann and I. G. Niemegeers, “An evaluation of location
management procedures,” in Proc. 3rd Annual Int. Conf. Universal
Personal Commun. (UPC’94), Sept./Oct. 1994, pp. 359-364.
[17] Y. Xiao, Y. Pan, and J. Lie, “Design and analysis of location manage-
ment for 3G cellular networks,” IEEE Trans. Parallel Distrib. Syst.,vol.
15, no. 4, pp. 339-349, April 2004.
[18] Y. Fang, “Movement-based mobility management and trade off analysis
for wireless mobile networks,” IEEE Trans. Computers, vol. 52, no. 6,
pp. 791-803, June 2003.
[19] R. Koodli and C. E. Perkins, “Fast handovers and context transfers in
mobile networks,” ACM Mobile Computing and Commun. Rev., vol. 31,
no. 5, Oct. 2001.
[20] J. McNair, I. F. Akyildiz, and M. D. Bender, “Handoffs for real-time
traffic in mobile IP version 6 networks,” in Proc. IEEE GLOBECOM,
vol. 6, Nov. 2001, pp. 3463-3467.
[21] W. K. Lai and J. C. Chiu, “Improving handoff performance in wireless
overlay networks by switching between two-layer IPv6 and one-layer
IPv6 addressing,” IEEE J. Select. Areas Commun., vol. 23, no. 11, pp.
2129-2137, Nov. 2005.
[22] N. Moore, “Optimistic duplicate address detection,” IETF RFC 4429,
April 2006.
Christian Makaya (S’06) received the M.Sc. de-
grees in Computer Science from University of
Montreal (2003) and in Telecommunications (2004)
from INRS-EMT, University of Quebec, Montreal,
Canada. He is currently pursuing a Ph.D. in Com-
puter Engineering (thesis carried out jointly with Er-
icsson Research Canada) at the Ecole Polytechnique
de Montreal, Montreal, Canada, while working as
a Graduate Research Assistant with the Mobile
Computing and Networking Research Laboratory
(LARIM). The current focus of his research interests
is on radio resource management, interworking architectures design, mobility
management, IP mobility, performance evaluation and quality of service (QoS)
provisioning for next-generation wireless networks.
Samuel Pierre (SM’97) received the B.Eng. degree
in Civil Engineering in 1981 from Ecole Poly-
technique de Montreal, Qu´
ebec, Canada, the B.Sc.
and M.Sc. degrees in Mathematics and Computer
Science in 1984 and 1985, respectively, from the
UQAM, Montreal, the M.Sc. degree in Economics
in 1987 from the University of Montreal, and the
Ph.D. degree in Electrical Engineering in 1991 from
Ecole Polytechnique de Montreal. From 1987 to
1998, he was a Professor at the University of Qu´
ebec
at Trois-Rivi`
eres prior to joining the T´
el´
e-Universit´
e
of Qu´
ebec, an Adjunct Professor at Universit´
e Laval, Ste-Foy, Qu´
ebec, an
Invited Professor at the Swiss Federal Institute of Technology, Lausanne,
Switzerland, and the the Universit´
e Paris 7, France. He is currently a Professor
of Computer Engineering at Ecole Polytechnique de Montreal, where he
is Director of the Mobile Computing and Networking Research Laboratory
(LARIM), Chairholder of the NSERC/Ericsson Chair in Next Generations
and Mobile Networking Systems, and the Director of the Mobile Computing
and Networking Research Group (GRIM). His research interests include
wireline and wireless networks, mobile computing, artificial intelligence, and
telelearning. Dr. Pierre is a Fellow of the Engineering Institute of Canada
(EIC). He is a Regional Editor of the Journal of Computer Science,an
Associate Editor of IEEE Communications Letters,IEEE Canadian Journal
of Electrical and Computer Engineering and IEEE Canadian Review,and
serves on the editorial board of Telematics and Informatics, edited by Elsevier
Science.
Authorized licensed use limited to: UNIVERSITI UTARA MALAYSIA. Downloaded on September 16, 2009 at 04:39 from IEEE Xplore. Restrictions apply.
