Beruflich Dokumente
Kultur Dokumente
(1)
T =
1, packet requests high throughput
0, otherwise
(2)
R =
1, packet requests high reliability
0, otherwise
(3)
The endpoint is the most knowledgeable component in the
network as it understands applications. Therefore MNs
identify the trafc type and estimate the QoS requirements
(e.g. delay and bandwidth guarantee) of the incoming ow.
Various methods maybe used for this purpose, such as
employment of packet inspection [15], however, this lies
beyond the scope of this paper.
For each incoming rst marked packet of a ow at AR m,
MN checks the DS eld bits, and then selects an MAP among
the available MAPs, which meets the ow QoS requirements.
Note only the delay sensitive ows are considered as HP
ows and the rest are considered as LP ows. The three
possible QoS requirements are
if [D T R] = [1 0 0], low delay is requested.
if [D T R] = [1 1 0], high throughput is requested.
if [D T R] = [1 0 1], high reliability is requested.
By knowing the D, T and R bits, the proposed MAP
selection mechanism (described in Table 2) is executed.
For the best-effort trafc, considered as LP class of trafc, a
load-balancing mechanism (described in Section 8) is
deployed, which works towards distributing the load evenly
among MAPs.
4.4.2 Extended router advertisement: Router
solicitations (RSs) and RAs help the MN to identify that it
has changed its subnet and to provide the MN with the
necessary information to congure a new CoA. A novel RA
protocol as an extension to the standard RA is introduced,
to include the network measurement information. It utilises
the bits in the reserved eld to disseminate MAPs current
load status. This eld is renamed as MAP utilisation eld,
and the protocol is referred to as the ex-RA. Fig. 2 depicts
the modied MAP option format.
Similar to MIPv6, ARs send ex-RA messages both on a
regular basis and in response to MNs requesting for them
through RS messages [7]. There is a tradeoff between
accurateness of MAP selection and the sorting cost of MAP
options by MNs. If the number of ex-RA is W
RA
, then the
sorting cost is regarded as W
RA
logW
RA
[16]. The more
often the ex-RA is sent, the more accurate the MAP
selection; however, the more sorting cost and power
consumption.
4.4.3 Signalling procedure for multi-MAP
registration: Fig. 3 shows how an MN registers with
more than one MAP simultaneously in a multi-MAP
domain HMIPv6 network.
When an MN connects to a new AR, it obtains an ex-RA
message, containing information on locally available MAPs
(e.g. load information, hop-distance, RCoA, Life time etc).
Then, BUs are invoked by the MN. BU in the multi-MAP
per domain environment is similar to that of the HMIPv6.
However, other than MAPs keeping binding information
about MNs LCoA, and RCoA in their BC, MNs and CNs
keep binding information of MNs RCoAs. HA and CNs
identify an MN with only one RCoA at any instant of time,
which is related to the MNs current supporting MAP [or
the primary MAP (P-MAP)]. The P-MAP is selected
through implementation of procedures explained in Sections
4.2, 4.3 and 4.4.
After the MAP selection stage, MN sends an LBU message
to the MAP, which binds its LCoA, with MAPs RCoA as
MNs P-MAP. The RCoA is registered with the HA and the
CN of the MN. The MN must not use one RCoA (e.g.
RCoA 1) derived from an MAPs prex (e.g. MAP 1) as a
CoA in its BU to another MAP (e.g. MAP 2). This would
force packets to be encapsulated several times (twice in this
example) on their path to the MN. Subsequently, the MN
also registers its LCoA with a secondary MAP (S-MAP)
and registers its S-RCoA with the HA and the CNs. A new
ag is added to the BU message to differentiate between
the primary and secondary RCoA [17]. The secondary
RCoA is used by the MN in the following cases:
change in capability of the P-MAP (e.g. when P-MAP
becomes overloaded);
MAP failure.
Fig. 2 MAP option format in proposed ex-RA
Table 2 Algorithm-i
1 If flow n arrives at AR mM
2 If D= 1
3 Find MAPs
4 where for each k
5 L
T
likn
, Ha
n
6 D
T
n
, Da
n
7
m[M
k
n
B
n
mk
.x
nm
.z
mk
+h
k
z
k
8 If R= 1
9 nd MAP k
10 (14)(18)
11 If T = 1
12 nd MAP k
13 (32)(35)
14 If T = 0 and
R= 0
15 nd MAP k
16 (28)(31)
17 End
18 End
19 End
20 Else if D= 0
21 Find an
MAP
22 (35)(38)
23
End
End
www.ietdl.org
IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186 179
doi: 10.1049/iet-net.2012.0141 & The Institution of Engineering and Technology 2014
When either of the above scenarios arises in an access
network, MN selects the S-MAP from its BC and sends
data through the S-RCoA to CNs. The CNs consider the
P-MAP has failed when they receive packets from S-RCoA.
Consequently, CNs update their BCs. The MN sends a BU
with the S-CoA to HA as soon as possible.
CN acts in a similar manner to the MNs procedure in the
mentioned scenarios. In such conditions, CN searches
through its BC and sends the data through the S-MAP (i.e.
the S-RCoA of the MAP was congured in advance and
stored in the CNs BC). If an MN receives packets from the
S-MAP, the MN considers that P-MAP is no longer used.
Then, the MN sends GBU to HA and to the CN to update
their BCs.
This mechanism steers IP ows from one MAP to another,
while providing uninterrupted services to MNs. Note that
only the non-real-time IP ows are redirected. As a result,
MNs do not generate long BU registration delays as their
LCoAs are registered with the S-RCoAs prior to the need
for use of the S-MAP.
5 Network model
An access network is dened as a given undirected graph G
(V, E), where V is the set of nodes and E is the set of links
interconnecting nodes. Let KV be the set of routers that
serve as MAPs, and R be the number of MAPs in the
access network. Let MV be the set of ARs in the network
and N V be the set of MNs. For a given AR m M, let
K
m
K be the set of MAPs that are advertised by ARm. Let
M
m
M be a set of ARs adjacent to AR m, and k K
m
be a
given MAP. Let
k
represent the capacity of each MAP k.
6 Analytic user mobility model
In this section, taking into account several parameters, a cost
function for total handover signalling delay cost is developed.
Handover latency is dened by many factors such as
movement detection delay (MDD), address conguration
delay (ACD), and the location update delay. A number of
solutions exist to minimise the MDD and ACD delays, to
negligible amounts [18]. This paper focuses on the cost
attributed to the location update signalling delay.
The cost is built assuming a single MAP hierarchy.
6.1 Location update cost
Based on the location update model derived in [19], location
update costs because of intra-domain and inter-domain
handovers are developed in this section.
6.1.1 Unit intra-domain handover location update
cost: The cost of sending a BU message and receiving a
binding acknowledgement from MAP k, during an
intra-domain handover, is directly proportional to the
hop-distance between the AR m and MAP represented by
h
AR
m
MAP
k
. Let and be the unit costs when a location
update procedure is performed in wireless and a wired link,
respectively. The unit intra-domain handover location
update cost between AR i and AR m located in an MAP
domain is dened as follows [19]
H
in
imk
= h + h
AR
m
MAP
k
v
2 (4)
6.1.2 Unit inter-domain handover location update
cost: Inter-domain handover is dened as the cost of
sending the GBU from an AR m to the GW, while
traversing MAP k, plus the hop-distance from the GW to
HA and CNs. The unit inter-domain handover location
update cost because of an inter-domain handover between
two ARs (AR i and AR m), allocated in different MAP
domains, is expressed as follows [19]
H
out
imk
= v. l
AR
m
MAP
k
+C
2 (5)
We dene C as a xed number of hops between the GW and
HA, and the CN, which is dened as
C = h
GWCN
+h
GWHA
(6)
In addition, l
AR
m
MAP
k
, is dened as follows
l
AR
m
MAP
k
= h
AR
m
MAP
k
+h
MAP
k
GW
(7)
Similarly, h
MAP
k
GW
represents the hop-distance between
MAP
k
and GW.
6.2 Probability of MNs performing handover and
MNs movement direction
Trafc is modelled as ow requests. Incoming ows are
assumed to arrive independently following Poisson
Fig. 3 Flowchart of MNs operation
www.ietdl.org
180 IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186
& The Institution of Engineering and Technology 2014 doi: 10.1049/iet-net.2012.0141
distribution with an average value . The ow holding time
and residence time are assumed to be exponential random
variables with mean and n minutes, respectively.
Fig. 4 illustrates the handover probability for a variety of
ow holding times and residence times. The smaller the
residence time, the higher the mobility speed. Also the
average handover probability of ows decreases by
decreasing the mean rate of incoming ows to the access
network. One of the MAP selection criteria is the handover
delay cost. Therefore the mobility pattern has a great impact
on performance of the proposed algorithm.
Similar to [11], the probability of an MN performing
handover and the MN movement direction, and the
internal and external costs of ARs are taken into account
to model the expected handover signalling overhead in the
network. Let
im
be the handover probability of an MN
from AR i to AR m, (referred to as the direction
probability of AR m), and let
i
be the probability of an
MN being attached to AR i (referred to as the demand
probability).
6.3 Handover signalling delay cost
In order to be able to express the handover signalling delay
cost in a mathematical programming setting, we dene the
following Boolean decision variables:
x
nm
=
1, flow n is connected to AR
m
0, otherwise
(8)
y
imk
=
1, if AR
i
and AR
j
are assigned
to MAP
k
0, otherwise
(9)
z
mk
=
1, AR
m
is assigned to MAP
k
0, otherwise
(10)
For each ow n attached to AR m, the total expected
intra-domain and inter-domain handover signalling costs
between two adjacent ARs i and m are the probability of
handover occurrence multiplied by the handover cost,
which is given as (4) and (5), respectively
L
in
im
=
k[k
m
j[M
i
H
in
imk
y
imk
x
ni
r
i
a
im
i [ M, n [ N
(11)
L
out
im
=
k[k
m
m[M
i
H
out
imk
1 y
imk
x
ni
r
i
a
im
i [ M, n [ N
(12)
In RH-MIPv6, when an MN attaches to an AR (AR
c
) that can
access more than one MAP, the probability of any of the
accessible MAPs to be selected follows a uniform
distribution. Then, MN registers with an MAP (MAP
c
). The
MN migrates to a new AR (AR
n
) coverage area according
to the direction probability of its AR
c
. If AR
n
can access
MAP
c
, then the handover is classied as an intra-domain
handover, otherwise an inter-domain handover. However,
by an intelligent selection of MAP, the handover delay can
be reduced considerably. In view of that, an MAP selection
algorithm is proposed in Table 2.
Let N
r
N represents a subset of ows with high reliability
requirements. For each n N
r
attached to AR i, the handover
signalling cost imposed by each MAP k accessible by AR i is
computed. The cost is sum cost of expected intra-domain and
inter-domain handover signalling costs, formulated in (11)
and (12), respectively
L
T
ikn
= L
in
im
+L
out
im
, k [ K
i
,
n [ N
r
i [ M, m [ M
i
(13)
For each ow nN
r
, the objective is to select an MAP, where
the handover signalling delay cost is minimised.
L
= min
k[K
i
L
T
ikn
(14)
Subject to
2y
imk
z
mk
+z
ik
, m, i [ M, k [ K (15)
k
z
mk
1, m [ V (16)
m[M
x
nm
= 1, n [ N
r
(17)
x
nm
, z
mk
, y
imk
[ 0, 1 { }, n [ N
r
i, m [ M, k [ K
(18)
Constraint (15) ensures that y
imk
can only take the value 1 if
both nodes i and m are assigned to MAP k, constraint (16)
ensures each AR is assigned to at least one MAP, whereas
constraint (17) ensures each ow is attached to one AR and
constraints (18) ensure that x
nm
, z
mk
and y
imk
are binary
values (e.g. either 0 or 1).
7 Packet delay cost
7.1 Average packet delay model in the access
network
The average packet delay cost consists of the delay costs of
the propagation, processing and queuing [6]. The queuing
delay can be deemed negligible, when the trafc load is Fig. 4 Hanover probability against residence time
www.ietdl.org
IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186 181
doi: 10.1049/iet-net.2012.0141 & The Institution of Engineering and Technology 2014
well below the capacity of the network (unloaded network).
When MAPs are deployed in the network topology,
creating bottlenecks, queuing needs to be explicitly taken in
consideration. The processing delay incurred by a network
entity, depends on its load status. We assume the
transmission delay cost is proportional to the distance
between the source and destination. The longer the distance
is, the larger is the round trip time experienced by MN.
7.2 Queuing delay cost
Using Kleinrocks independence approximation [20], each
link/edge can be modelled as an M/M/1 queue. It is
assumed that bottlenecks are at MAPs in the network.
Let B
n
mk
be the bandwidth request associating with ow n,
attached to AR m and traversing MAP k. We dene the total
bandwidth utilisation of MAP k, as follows
h
k
=
m[M
n[N
B
n
mk
x
nm
z
mk
_ _
, k [ K
m
(19)
By Littles law, the packet queuing delay cost in MAP k is
modelled as follows
P
APQ
k
=
1
6
k
h
k
(20)
Hence, the packet queuing delay cost at MAP k experienced
by ow n, connected to AR m, is as follows
P
APQ
nk
=
1
6
k
h
k
x
nm
z
mk
i [ V, k [ K
m
(21)
7.3 Transmission and the processing delay costs
In this section, the transmission and the processing delay
costs at each network entity such as HA and MAP are
developed, based on the model derived in [21]. The sum
transmission and the processing delay cost (referred to as
the packet delivery cost) incurred from the CN to an MAP
is presented as follows
D
CNMAP
= l b h
CNHA
+h
HAMAP
_ _
+l E(S) 1
_ _
b h
CNMAP
+l P
HA
(22)
where h
i j
represents the distance between the two entities of
i and j in the network and E(S) is the average ow size (in the
unit packet). In addition, is the unit transmission cost in a
wired link and P
HA
denotes the processing cost at the HA.
Similarly, the packet delivery cost from an MAP to AR D
MAP AR
is as follows
D
MAPAR
= l E(S) P
MAP
b h
ARMAP
_ _
(23)
where P
MAP
is the processing cost of the MAP, including a
lookup and a packet encapsulation/decapsulation costs. The
lookup cost of an MAP is assumed to be proportional to the
logarithm of the number of ows managed by that MAP
[21] and the packet encapsulation/decapsulation cost is a
constant value. Hence, the processing cost of MAP k is as
follows
P
MAP
k
= d log
m[M
k
x
nm
_ _
+O (24)
is a weighing factor and O is the encapsulation and
decapsulation cost.
The last component is the packet delivery cost in the
wireless link between AR and MN, denoted by D
ARMN
.
D
ARMN
= l E(S) j (25)
where is the unit transmission cost in wireless link.
Let N
d
N represents a subset of ows with low delay
tolerability. Hence, the average packet delivery cost of
MAP k experienced by ow n, connected to AR m, is as
follows
D
total
nk
= D
CNMAP
k
+D
MAPAR
m
_
+D
AR
m
MN
n
_
x
nm
z
mk
,
k [ K
m
(26)
Using (26) and (21), the packet delay cost at MAP k for ow
n, connected to AR m, is given as
D
T
n
= D
total
nk
+D
APQ
nk
, i [ V, k [ K
m
, n [ N
d
(27)
For each ow n N
d
, the objective is to select an MAP, where
the packet delay cost is minimised
D
=
n[N
d
, k[K
m
D
T
n
_ _
(28)
subject to
k
z
mk
1, m [ V (29)
m[M
x
nm
= 1, n [ N
d
(30)
x
nm
, z
mk
, y
imk
[ {0, 1}, n [ N
d
i, m [ M, k [ K
(31)
Constraint (29) ensures each AR is assigned to at least one
MAP, constraint (30) ensures each ow is attached to one
AR, and constraints, whereas constraint (31) ensures that
x
nm
, z
mk
and y
imk
are binary values (e.g. either 0 or 1).
8 Flows with high-throughput requirement
In this section, a linear problem formulation of selecting the
least utilised MAP for a ow with high-throughput
requirement is provided. Let N
t
N represents a subset of
ows with high-throughput requirements and B
n
mk
be the
bandwidth request associated with ow n, attached to AR
m. The least utilised MAP based on (19), is given as follows:
h
= min
k[K
m
h
k
_ _
(32)
www.ietdl.org
182 IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186
& The Institution of Engineering and Technology 2014 doi: 10.1049/iet-net.2012.0141
subject to
m[M
k
n
B
n
mk
x
nm
z
mk
+h
k
6
k
(33)
B
n
mk
0, n [ N, m [ V, k [ K (34)
x
nm
, z
mk
, [ 0, 1 { }, n [ N
t
, m [ M, k [ K (35)
Constraint (33) ensures MAP capacity is satised for all
MAPs, constraint (34) ensures bandwidth requirement of
ows are non-zero and constraint (35) ensures that x
nm
and
z
mk
are binary values (e.g. either 0 or 1).
9 Load balancing model
An MAP is congested when its total bandwidth utilisation
exceeds the bandwidth utilisation threshold . The threshold
is dened as 80% of the MAPs total capacity. Assuming
that 80% is the expected trafc percentage.
In the proposed algorithm, for HP ows, MAPs are
selected to satisfy the ows QoS requirements, whereas a
new MAP selection mechanism is proposed for LP trafc
ows. The objective is for each LP ow, an MAP is
selected to provide better load balance within network.
Based on the load balance model derived in [22], we
develop a balance criterion as follows:
An MAP domain k is balanced if all MAP domains satisfy
the balance criterion
h
low
h
k
h
up
(36)
where
h
low
= h
total
/R
1 w
(37)
h
up
= h
total
/R
1 +w
(38)
Also
total
is dened as follows
h
total
=
k
h
k
, k [ K (39)
total
represents the total bandwidth request in the network.
low
and
up
represent the lower and upper bandwidth
utilisation thresholds for each MAP domain, respectively
and is a parameter satisfying 0 < < 1. Constraints (37)
and (38) specify a range for MAP k, within which the MAP
is accepted as balanced. The smaller the value, the tighter
the constraint is. The value 0.1 is used for in our
implementation as in the similar work [22]. Therefore for
each LP ow, MN selects an MAP, which provides the best
load balance in the network.
10 Proposed algorithm
Every time an MN enters an access network, it receives an
ex-RA (introduced in Section 4.4.2) from its current point
of attachment and stores the received MAP option(s) of the
available MAPs in its MAP list. The MAP list consists of
the dist eld, MAP utilisation eld (introduced in
Section 4.4.3).
MNs extract the necessary information received in MAP
options for available MAPs (e.g. hop-distance and the MAP
trafc load status) and determine the handover delay cost
(formulated by (13) in Section 6.3) and the packet delay
cost (dened in (26) in Section 7.3) imposed by each
available MAP.
MNs identify the trafc type and estimate the QoS
requirements of the incoming ow and use delivery priority
values to differentiate between the packets generated by
them (explained in Section 4.4.1). By knowing the D, T
and R bits (the fourth, fth and sixth bits of the DSCP in
the IP header), the proposed MAP selection mechanism
(described in Table 2) is executed. There are three possible
QoS requirements, (i) low-delay requirement, (ii) high
throughput requirement and (iii) high reliability
requirement. Line 1: in the proposed algorithm, for HP
ows, MAPs are selected to satisfy the ows QoS
requirements. Lines 2, 3, 4, 5, 6 and 7: let D
an
and H
an
represent the minimum acceptable bound for required
handover delay and packet delay for ow n, respectively.
Using this information, for each ow n, if ow is delay
sensitive, the algorithm selects a serving MAP to satisfy the
ows QoS. Therefore MAPs that meet the acceptable
handover delay (line 5), packet delay (line 6) requirements
of the ow and also meet the MAP capacity constraint (line
7) are selected. The MAP capacity constraints ensure that
the total amount of trafc load going through each MAP is
not greater than the capacity of that MAP. Then
lines 8, 9 and 10: if ow requires high reliability, the MAP
with minimum handover delay cost is selected;
lines 11, 12 and 13: if ow requires high throughput, the
least utilised MAP is selected; and
lines 14, 15 and 16: if ow requires high reliability, the
MAP with minimum packet delay cost is selected.
Lines 20, 21 and 22: for the best-effort trafc, considered
as LP class of trafc, a load-balancing mechanism is
deployed, which selects an MAP to meet the balance
criteria dened in (36) and works towards distributing the
load evenly among MAPs.
Table 2 shows how the most suitable MAP is selected.
11 Simulation setup and evaluation
A simulation-based study is developed using MATLAB,
which supports hierarchical mobile IP architecture. Fig. 5
outlines the network simulation topology. Dashed lines
show possible user movements between ARs, and solid
lines present wired links between routers. The ARs are
connected to MAPs through intermediate routers, having
point to point wired links, with 10 ms delay allocated to
each link. The MAPs are connected to HA and CN via
wired network. The choice of having a one layer MAP
hierarchy is to focus the research on the overlapping MAP
domains in a single hierarchy level.
By expanding the size of the MAP domain overlapping
regions, the trafc ows are more evenly distributed in the
network. This is achieved by making the residual capacity
of the lightly loaded MAPs available to ARs located in the
overlapped regions of MAP domains. Implementations
network partitioning in real-time requires the size of
overlapping regions between MAP coverage areas to
dynamically shrink or expand depending on MNs mobility
parameters, addressed in details in [11]. However, this lies
beyond the scope of this paper. Over expanding the MAP
domains over the neighbouring domains increases the
intra-domain handover rate considerably. Hence, the total
www.ietdl.org
IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186 183
doi: 10.1049/iet-net.2012.0141 & The Institution of Engineering and Technology 2014
handover signalling overhead in the network increases as a
result of overlap formation instead of being decreased. In
addition, as the number of ARs located in the overlapping
region expands; the residual capacities of the MAPs to
which ARs are assigned to is shared by the new AR(s).
Hence, reductions in load concentration on the congested
MAPs are achieved with the cost of an increase in trafc
load on the new MAP. However, once MAPs reach their
maximum capacity usage, further expansion of overlapped
regions between MAP domains has no longer an impact on
congestion reduction of bottlenecks. Owing to these
remarks in a single MAP hierarchy, in our simulation a
xed 50% overlap size between MAP domains is permitted.
Consequently, the focus is to evaluate the performance of
the proposed intelligent multi-MAP registration algorithm
(in network with overlapping MAP domains) in this paper
and the algorithm proposed in [13], which does not
consider QoS of ows in process of MAP registration.
The capacity of each MAP is set to 3 Mbps. The ow
bandwidth requests are uniformly distributed within the
interval of (100500 KBps), representing from web to video
applications. The ows are classied into six different
trafc class types. Then, three different bits of D, R and T
are assigned to each ow indicating their corresponding
class types. Table 3 shows the used parameters for packet
delay cost measurements.
The average rate of Poisson distributed incoming ows is
set as = 10. The means of exponentially distributed ows
holding time and residence time are set as = 20 min and
n = 5 min, respectively.
Figs. 6a and b present the percentage of bandwidth
consumption of MAPs over a certain period of time. In
Fig. 6a, the ow requests enter the access network, and
become distributed across MAPs according to the non-QoS
aware multi-MAP registration scheme proposed in [13].
Fig. 6b illustrates the distribution of ow requests across the
network MAPs according to the proposed algorithm
illustrated in Table 2. The simulations begin at t = 0s.
Fig. 6a shows that MAP 1 and MAP 2 are constantly
selected and reach the 80% utilisation threshold, at t = 25s
and t = 45s, respectively. Therefore the non-optimal
selection of MAPs leads to rapid increase in bandwidth
utilisation of these two MAPs, and makes them points of
bandwidth aggregation in the network, whereas MAP 3 and
MAP 4 stay underutilised.
It is evident in Fig. 6b that the maximum utilisation of
MAPs in the network remains at 65% of the total MAP
capacity and does not reach their maximum capacities. This
is because of accessibility of trafc ows to more MAP
resources (capacity) for ARs located in the overlapped
region of MAP domains. In addition, severe bottleneck
congestion around MAPs is mitigated, hence enhancing the
network performance in terms of increasing network
throughput. Fig. 6b also illustrates that the trafc load is
more uniformly distributed among MAPs as a result of the
Table 3 System parameters
O E(S) P
HA
1.0 2.0 0.5 2.0 10 40
Fig. 5 Simulation network topology
Fig. 6
a MAP bandwidth utilisation in RH-MIPv6
b Bandwidth utilisation, with the proposed algorithm in use
www.ietdl.org
184 IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186
& The Institution of Engineering and Technology 2014 doi: 10.1049/iet-net.2012.0141
proposed algorithm implementation in the network, than by
implementation of the proposed algorithm in [13] (Fig. 6a).
Fig. 7 shows as the trafc demand increases in the
overlapping MAP domain environment, more number of
ows are admitted to the network and network throughput
increases in turn. Therefore it is clearly evident that the
proposed algorithm performs more efciently by a
considerable margin, than the one proposed in [13] by
increasing the mean satised bandwidth demands of ows
by maximum of 74%.
In order to evaluate the impact of proposed load-balancing
algorithm in terms of users perceived performance of the
network, the amount of bandwidth rejection is measured
against time. Fig. 8 illustrated the total bandwidth rejection
in the network within a specic duration of time.
Fig. 8 illustrates that the total bandwidth rejected is the
network increases as more ows arrive to the network. It
also shows the total bandwidth rejected in the network is
considerably higher in [13], than that of caused by the
employment of proposed algorithm. The weak performance
of RH-MIPv6 is a result of a random MAP selection;
hence, an uneven distribution of trafc between MAPs
(which is evident in Fig. 6a). Additionally, with no
overlapping regions between MAP domains, the trafc load
initiated from ARs located in an MAP domain is restricted
to ow through only that MAP. Consequently, the MAP
becomes congested and no more ows are accommodated
by that MAP, and the new ows are rejected. The reduction
in average bandwidth rejection achieved by the proposed
algorithm is because of the shift of LP ows from the
highly utilised MAPs to the lightly loaded MAPs. The
proposed scheme selects alternative MAPs for LP ows
rather than forcing all trafc ows through the same MAP
(i.e. as proposed in [13]) to ensure a uniform load
distribution between MAPs. Thus, employing the proposed
algorithm in the network yields 71% reduction in total
bandwidth rejection in the network.
12 Conclusions
In this paper, the effect of overlapping domain regions of
consecutive MAPs on the trafc distribution and the
degree of load balance between MAPs is studied.
Introduction of overlapping MAP domains in the same
hierarchy level of HMIPv6-based network architecture
enforces the need for multiple registrations of MAPs for
MNs. This paper proposes an adaptive QoS aware
multiple MAP registration algorithm, which separates the
selection scheme for the HP and LP ows. For each HP
ow, the proposed algorithm selects the most suitable
MAP based on QoS requirements of the ow, whereas for
each LP ow an explicit MAP selection is employed,
which provides the best balance, in terms of bandwidth
utilisation among MAPs. The simulation results illustrate
that implementing the proposed algorithm in the novel
network architecture, provides more available resources for
HP ows, in comparisons with the multi-MAP registration
proposed in [13], in non-overlapping MAP domain access
networks. Accordingly, a maximum of 71% drop in
total ow bandwidth rejection is obtained. In addition, the
mean satised ow bandwidth demand is increased by a
maximum of 74%. This improvement is because of shift
of LP ows from the heavily loaded MAPs to more
lightly loaded ones. The trafc shift also has a
load-balancing impact among MAPs. Consequently,
congestion in the network decreases, whereas QoS of HP
ows are satised.
13 Acknowledgment
The authors thank Alexandre Jaron for his helpful
suggestions.
14 References
1 Johnson, D., Perkins, C., Arkko, J.: Mobility support in IPV6, IETF
RFC3775, June 2004
2 Yan, Z.-W., Zhou, H.-C., Zhang, H.-K., Zhang, S., Chao, H.-C.: Design
and implementation of PMIPv6 based multi homing for
make-before-break handover, J. Internet Technol., 2010, 11, (1),
pp. 110
3 Lee, J.-H., Ernst, T., Deng, D.-J., Chao, H.-C.: Improved PMIPv6
handover procedure for consumer multicast trafc, IET Commun.,
2011, 5, (15), pp. 21492156
4 Han, Y.-H., Min, S.-G.: Performance analysis of hierarchical mobile
IPv6: does it improve mobile IPv6 in terms of handover speed, Wirel.
Pers. Commun., 2009, 48, (4), pp. 463483
Fig. 7 Mean MAP bandwidth utilisation against time
Fig. 8 Bandwidth rejection against time
www.ietdl.org
IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186 185
doi: 10.1049/iet-net.2012.0141 & The Institution of Engineering and Technology 2014
5 Leung, G.K.: Proxy mobile ipv6, RFC 5213, August 2008], [Soliman
H., Castelluccia C., Malk K.E., Bellier L.: Hierarchical mobile IPv6
mobility management (HMIPv6). IETF RFC 414, August 2005
6 Dev Pragad, A., Friderikos, V., Pangalos, P., Aghvami, H.: The impact
of mobility agent based micro mobility on the capacity of wireless access
networks (IEEE GLOBCOM, 2007)
7 Soliman, H., Castelluccia, C., Malk, K.E., Bellier, L.: Hierarchical mobile
IPv6 mobility management (HMIPv6).IETF RFC 414, August 2005
8 Pack, S., Nam, M., Kwon, T., Choi, Y.: An adaptive mobility anchor
point selection scheme in hierarchical Mobile IPv6 networks,
Comput. Commun., 2006, 29, (16), pp. 30663078
9 Natalizio, E., Scicchitano, A., Marano, S.: Mobility anchor point
selection based on user mobility in HMIPv6 integrated with fast
handover mechanism. Proc. IEEE WCNC 2005, March 2005, Vol. 3
10 Wang, L., Gaabab, B., Binet, D., Kofman, D.: Novel MAP selection
scheme using location history in hierarchical MIPv6 networks (IEEE
WCNC, 2008)
11 Mirsayar, A., Barkoosaraei, A., Aghvami, H.: Intelligent overlapping
MAP domain forming for mobility management in HMIPv6 access
networks (IEEE WCNC, 2012)
12 Mirsayar, A., Barkoosaraei, A., Aghvami, H.: Dynamic partitioning of
IP-based wireless access networks. Elsevier Computer Networks, 2012.
Available at http://dx.doi.org/10.1016/j.comnet.2012.09.005
13 You, T., Pack, S., Choi, Y.: Robust hierarchical mobile IPv6
(RH-MIPv6) an enhancement for survivability & fault-tolerance in
mobile IP systems (VTC, 2003)
14 Nichols, K., Blake, S., Baler F., Black, D.: Denition of the
differentiated services eld. IETF RFC2474, December 1998
15 Allot Communications, Allot Resource Center: Digging deeper into
deep packet inspection (DPI). December, 2007. Available at www.
dpacket.org/articles/digging-deeper-deep-packet-inspection-dpi
16 Kim, C.: Convergence in broadband and mobile networking:
international conference. ICOIN 2005, Proc.: Google eBook, Jeju
Island, Korea, January 31February 2, 2005
17 Thing, V., Lee, H., Xu, Y.: Designs and analysis of local mobility
agents discovery, selection and failure detection for mobile IPv6.
Proc. IEEE MWCN, September 2002
18 Jeong, C.S., Song, Y.M., Jo, S.U.: Dynamic iterative method for fast
network partitioning. Proc.: Eighth Int. Conf., HPCN Europe, 2000,
pp. 8188
19 Pragad Audsin, D., Friderikos, V., Hamid Aghvami, A.: Optimal
conguration of mobility agents in broadband wireless access
networks. IEEE GLOBCOM Workshop on BWA, 2008
20 Kleinrock, L.: Communication nets: stochastic message owand delay
(Dover Publications, 1964) ISBN: 0486611051
21 Pack, S., Nam, M., Choi, Y.: A study on optimal hierarchy in
multi-level hierarchical mobile IPv6 networks (IEEE GLOBCOM,
2004)
22 Apostolopoulos, G., Gudrin, R., Kamat, S., Tripathi, S.K.: Quality of
service based routing: a performance perspective. Conf. the ACM
Special Interest Group on Data Communication (SIGCOMM),
Vancouver, B.C., 1998
www.ietdl.org
186 IET Netw., 2014, Vol. 3, Iss. 3, pp. 176186
& The Institution of Engineering and Technology 2014 doi: 10.1049/iet-net.2012.0141