Integration of Mobile IP and Multi-Protocol Label Switching

Zhong Ren Chen-Khong Tham Chun-Choong Foo Chi-Chung Ko

Department of Electrical and Computer Engineering
National University of Singapore
10 Kent Ridge Crescent, Singapore 119260
Tel: (65) 874 5095, Fax: (65) 779 1103
E-mail: engp9021, eletck, engp7643, elekocc@nus.edu.sg
AbstractMulti-Protocol Label Switching is a technology that combines
the simplicity of IP routing with the high-speed switching of ATM. Mobile
IP is a protocol that allows users to move around and yet maintain continuous IP network connectivity. In this paper, we propose a scheme to integrate
the Mobile IP and MPLS protocols. The integration improves the scalability of the Mobile IP data forwarding process by leveraging on the features
of MPLS which are fast switching, small state maintenance and high scalability. In addition, we have removed the need for IP-in-IP tunneling from
HA to FA under this scheme. This paper covers some issues regarding Mobile IP scalability and also defines the signaling and control mechanisms
required to integrate MPLS and Mobile IP.
KeywordsMulti-protocol Label Switching (MPLS),
Label Distribution Protocol (LDP), Mobile IP (MIP)

I. I NTRODUCTION
MPLS is a packet forwarding scheme. Since labels have only
local significance between two adjacent LSRs on a route, MPLS
has high scalability. Mobile IP is designed to support mobile
computing over the Internet.
Currently there are proposals to incorporate IP-based technologies into the core networks of future wireless cellular systems such as Universal Mobile Telecommunications System
(UMTS) [1], Iceberg Project [2] and Cellular IP [3]. Mobile IP
could potentially provide host mobility solution in these future
networks. Since the number of users and terminals connected to
these future systems would be very large, the scalability of the
Mobile IP solution is of great concern and interest. There have
also been work in integrating ATM as the transport provider into
these core networks [4]. Since MPLS and ATM are very closely
related, it would be desirable to incorporate MPLS into these
core networks too.
In this paper, we propose a scheme to integrate the Mobile
IP and MPLS protocols. The integration improves the scalability of the Mobile IP data forwarding process. Our work here
paves the way for the incorporation of both the Mobile IP and
MPLS protocols into these future IP-based core networks, and
also provide mobility support for MPLS.
The organization of the rest of the article is as follows. Section II briefly presents the basics of MPLS. Section III gives a
short introduction to the Mobile IP basic operation scheme. Section IV presents the potential scalability problem of Mobile IP.
In Section V, we present our solution to integrate MPLS into
Mobile IP in details. Evaluation results are presented in Section
VI. Finally, our conclusion is presented in Section VII.
II. M ULTI - PROTOCOL L ABEL S WITCHING
MPLS is a technology that integrates the label-swapping
paradigm with network-layer routing [5]. Each MPLS packet

In
Port
In
Lbl

L2 Header

Label

L3 Header

Address
Prefix


Out
port

Out
Lbl

128.197

18.181

164.67

......

Out
Port

Out
Lbl

In
Lbl


...

...

In
Lbl

Address
Prefix

Out
Port

128.197

18.181

9
6

...

In
Port

...

...


e.g.
IPv6

e.g.
ATM

Label

COS

TTL

20

0


18.181.0.31

L2 Header Label/ IPHeader L3 Data

Data

LSR2

LSR1
4

...

5
...

0
...

...
1

LSR3

Shim
header

......

Out
Lbl

18.181.0.31

9 18.181.0.31


18.181.0.31

Data

Data

Data

MPLS
Backbone

Label/ L2 Header

LSR4

L3 Data

Figure 1. Label Encoding

Figure 2. MPLS Operation Procedure

has a label. Depending on different Layer 2 and Layer 3 technologies involved, different label encoding schemes can be used.
They are illustrated in Figure 1.
Label swapping is done by associating labels with routes and
using the label value in the packet forwarding process. Packets
are classified and routed at the ingress Label Switching Routers
(LSRs) of an MPLS-capable domain. The mapping between IP
packets and a Label Switched Path (LSP) is done by providing
a Forwarding Equivalence Class (FEC) specification for each
LSP. MPLS labels are then inserted. When an LSR receives a
labeled packet, it will use the label as an index to look up the
forwarding table. This is faster than the process of parsing the
routing table and search for the longest match done in IP routing. The packet is processed as specified by the forwarding table
entry. The incoming label is replaced by the outgoing label, and
the packet is switched to the next LSR. Before a packet leaves
an MPLS domain, its MPLS label is removed. The MPLS operation procedure in a sample network is shown in Figure 2.
MPLS uses the Label Distribution Protocol (LDP) [6] to distribute the labels and set up LSPs. LSP setup can be traffic,
request or topology-driven. In the case of a topology-driven
scheme the labels are pre-assigned according to existing routing protocol information.
III. M OBILE IP
Mobile IP is a protocol to support mobile computing over the
Internet. A Mobile IP scheme has been adopted by the IETF for
standardization in IP version 4 (IPv4) [7]. A Mobile Node (MN)
is identified by the IP address it has when it is in its home network, called its home address. When a MN moves away from its
home network to a foreign network, it obtains a temporary CareOf-Address (COA) from the Foreign Agent (FA) in the foreign
network. The MN registers with a Home Agent (HA), which is
typically a router, in its home network, informing the latter of
its COA. Any Correspondent Node (CN) wishing to communi-

MN

FA


Foreign Agent

Global
Interne
t

Mobile
Node

LSR1

LSR3

2 1

Home Agent

LSR2

MN

FA

Figure 3.
Flow

CN1

MPLS
Network

3 1

LSR4

Modify Routing

HA

MN


FA

HA

Datagram
Looks Up
Label Table

Table
MIP registration
request

IP Host

Mobile IP Datagram

CN

MIP registration
request

MN


HA

MIP agent
advertisement

Datagram

LSP Setup

Figure 4. Sample Architecture

Looks Up
Label Table

Modify Label
Table

cate with the MN need not be aware that the MN has moved; it
simply sends IP packets addressed to the MNs home address.
These packets are routed via normal IP routing to the MNs
home network, where they are intercepted by the HA. The latter
encapsulates each such packet in another IP packet which contains the MNs COA as destination address. Thus these packets
are delivered to the MNs new location by a tunneling process.
Figure 3 illustrates the routing of datagram to and from a MN
away from home.
IV. M OBILE IP S CALABILITY I SSUES
The operation of Mobile IP involves three different activities,
which are the agent advertisement process, the registration process and the data forwarding process. It is crucial that these three
different activities operate efficiently in order for the Mobile IP
protocol to be scalable to systems consisting of huge numbers
of mobile hosts.
The data forwarding process of a Mobile IP HA works as follows. For every IP packet that the HA receives, it needs to check
if the destination IP address of the packet matches any MNs that
are currently registered in a foreign network. If yes, the HA will
perform IP tunneling of the packet by adding an IP header to
the packet and then sending it out to the routing process for forwarding. If no match is found, the HA just sends the packet out
to the routing process for forwarding.
The amount of processing required by the HA in this forwarding process depends on the number of MNs belonging to the
home network that are currently registered in a foreign network.
If there are many such kind of MNs, the forwarding process will
take very long. Considering that every packet forwarded by the
HA has to undergo this forwarding process, the overhead of this
packet forwarding process may be too high even after optimization through the use of appropriate data structures and lookup
algorithms [8]. This poses a scalability concern that affects the
use of the Mobile IP protocol in future wireless mobile systems.
V. MPLS AND M OBILE IP
In this section we will present our solution to integrate both
MPLS and Mobile IP in details.
A. Single MPLS Domain
A.1 Architecture
As shown in Figure 4, HA and FA are edge LSRs and belong
to the same MPLS domain. They support both MPLS and Mobile IP functionality. We assume that the MN home address is
a.b.c.d and the HA address is a.b.c.e. In addition, we assume
that FA COA is w.x.y.z.

MIP registration
reply

MIP registration
reply

Figure 5. Registration Procedure

Looks Up
Routing Table


Datagram

Figure 6. Datagram Delivery

A.2 Registration Procedure

1. MN determines whether it is at home when it receives agent
advertisement messages broadcast by the mobility agents.
2. If the MN is in a foreign network, it acquires a temporary
COA from FA and sends registration request to FA.
3. Since FA is an edge LSR, it will analyze the incoming registration request message and get the destination address of it.
4. Then FA updates its routing table and adds a host specific
row with the value of MN home address. In addition, it sets the
outgoing port value of this entry to be the incoming port number
from which it received the registration request.
5. Based on the IP routing table, FA forwards the registration
request message toward HA.
6. The packet is forwarded to HA hop-by-hop using IP routing.
7. When HA gets the registration request message and learns
the COA, it searches its label table to find the row with the MN
home address as FEC. The second row in Table I is that one.
8. Then, it will send a label request using LDP to FA with the
COA as FEC. FA replies with an LDP label mapping message to
HA. When this label mapping message arrives at HA, the LSP
would have been established (the first row in Table I is created
by LDP). In the case of the topology-driven scheme, the best
effort LSPs from FA to HA and from HA to FA would have
already been established using conventional IP routing. So, for
best effort traffic, we can use that best effort LSP in order to
reduce the registration time.
9. HA changes the row in its label table that uses the MN home
address as FEC. It sets the empty out label and outgoing port
entries to the values of out label and outgoing port of the LSP
from HA to FA. In this way, HA can relay the packets destined to
MN home address to its current location in the foreign network.
10. After that, HA sends a registration reply to FA along the
LSP from HA to FA.
11. When FA receives the registration reply, it records the incoming port number and in label value of the reply message.
Then it adds a new row in its label table. Table ?? illustrates the
example label table of FA after receiving the registration reply.
Figure 5 illustrates the procedure of Mobile IP registration.
Table I is an example label table of HA after registration. The
out label value and outgoing port number of LSP from HA to
FA are 5 and 1 respectively. The first row of Table I is the label
binding for the LSP from HA to FA. Since HA is the ingress
LSR, the in label value entry is empty. The second row is the la-

bel binding for the LSP from CN to MN. Since HA is the egress
LSR of this LSP, originally the outgoing port and out label entries are both empty. But HA has set these two entries to the
values of out label and outgoing port of the LSP from HA to FA
after receiving registration request.
TABLE I
E XAMPLE L ABEL TABLE OF HA
Incoming
Port
2
1
...

In
Label
9
...

FEC
w.x.y.z
a.b.c.d
...

AFTER

R EGISTRATION

Outgoing
Port
1
1
...

Out
Label
5
5
...

A.3 Datagram Delivery

1. Packets from a CN to the MN are addressed to the MN home
address and intercepted by the HA.
2. HA uses the incoming label value as an index to look up its
label table. According to Table I, HA inserts the label value in
the second row of the label table into packet and sends it out
through the port indicated in the same row. If MN is still in the
home network, out label and outgoing port entries are empty.
The packet will be sent to the IP layer and sent out based on the
corresponding routing table entry to MN.
3. The packet is delivered from HA to FA along the LSP by
label swapping.
4. FA receives the packet and looks up its label table. Since it is
the egress of the LSP from HA to FA, the out label and outgoing
port entries are empty. FA strips off the label and sends the
packet to the IP layer.
5. Finally, FA forwards the packet to MN based on the information in the newly-added host specific row of the routing table.
6. MN receives the packet sent by CN.
Figure 6 illustrates the procedure of Mobile IP registration.
As noted above, integrating MPLS and Mobile IP makes IPin-IP tunneling unnecessary in the data forwarding process. Instead we use MPLS to switch the packet to the foreign network.
Switching is much faster than conventional IP forwarding. The
whole forwarding process is done at the MPLS layer and HA
doesnt need to involve the IP layer. This improves the scalability of the Mobile IP protocol. In addition, since label header is
much smaller than IP header, the traffic overhead from HA to FA
is also reduced. Moreover, with Constraint-Based Label Distribution Protocol (CR-LDP) [9] we can setup an LSP satisfying
the QoS requirements of the traffic and do traffic engineering
[10].

2
New FA

MN

2


1
LSR1

21
LSR2

1
LSR3

LSR2

1


LSR4

HA

MPLS
Network

1
LSR6

MN

LSR3

MN
FA

3 1

CN1

3 1

CN1

21
LSR1

Old FA

2
LSR5

CN2

MN

LSR4

HA

MPLS
Network

MN

CN3

Figure 7. Multiple FA Sample Architecture

Figure 8. Multiple CN Sample Architecture

to the corresponding values of the third one so that the packets

destined to MN home address can be relayed to the new foreign
network. At the New FA, it adds a host specific row with the
value of MN home address in its routing table.
TABLE II
E XAMPLE L ABEL TABLE OF HA AFTER MN M OVES TO A N EW F OREIGN
N ETWORK
Incoming
Port
2
1
2
...

In
Label
9
...

FEC
w.x.y.z
a.b.c.d
a.s.d.f
...

Outgoing
Port
1
1
1
...

Out
Label
5
6
6
...

3. The datagram delivery procedure described in the previous

section is repeated once between the HA and new FA.
4. Finally MN receives the packet sent by CN.
A.5 Mobile Node Homing
In this section we will discuss the registration and data delivery schemes for MN movement from the foreign network back
to its home network:
1. MN finds it is back to home network after receiving agent
advertisement messages broadcast by its home agent.
2. It sends a deregistration request message to the home agent
with registration lifetime field equal to zero. The COA in this
message is the COA of the HA.
3. HA deletes the out label value and outgoing port number
from the second row of its label table that are added during the
last registration with the FA. As illustrated in Table III, these
two entries are left empty.
4. When packets destined to MN home address arrive at HA, it
strips off the label and sends the packets to the IP layer.
5. Then it searches the IP routing table. The packets are sent
out to MN based directly on the information in the routing table.
6. MN receives the packet sent by CN.

A.4 Multiple Foreign Agents

In this section, we will discuss the registration and data delivery schemes for MN movement from one FA to another FA. We
assume that the new FA COA is a.s.d.f.
Once the MN moves to a new FA:
1. The registration procedure described in the previous section
is repeated once between the HA and new FA.
2. After registration, the third row in Table II is new: it is the
label binding for the LSP from HA to new FA. The outgoing
port number and out label value in the second row are changed

TABLE III
E XAMPLE L ABEL TABLE OF HA AFTER MN M OVES BACK TO H OME
N ETWORK
Incoming
Port
2
1
...

In
Label
9
...

FEC
w.x.y.z
a.b.c.d
...

Outgoing
Port
1
...

Out
Label
5
...

A.6 Multiple Correspondent Nodes

B.1 Multiple MPLS Domains

If there are multiple CNs communicating with MN, once the

MN moves to a foreign network, all traffic from those CNs need
to be relayed to the current location of MN. In Figure 8, there
are three CNs communicating with the MN.
In best effort case, the registration procedure is the same as
the procedure in the case of one CN. The sample label table of
HA after registration is illustrated in Table IV. The first row is
the label binding for the LSP from HA to FA. The second, third
and fourth rows are the label bindings for the LSPs from each
CN to HA. Outgoing port and out label in these rows have been
set to the out label value and outgoing port number of the first
row. So packets arriving along these three LSPs are all sent out
from the same port with the same label value to foreign network.

As shown in Figure 9, HA and FA are edge LSRs and belong

to two different MPLS domains which are directly connected.
They support both MPLS and Mobile IP functionality.
Here the two edge LSRs (LSR3 and LSR5) are LDP Border
Gateway Protocol (BGP) peers [12]. That means they can exchange label information between them. So in this case, we can
establish a LSP from HA to FA across the link connecting these
two different MPLS domains. Our registration and data delivery
schemes described in previous sections can be used here without
any modification.

TABLE IV
E XAMPLE L ABEL TABLE OF HA AFTER R EGISTRATION IN M ULTIPLE CN
C ASE
Incoming
Port
2
1
1
1
...

In
Label
9
8
7
...

FEC

Outgoing
Port
1
1
1
1
...

w.x.y.z
a.b.c.d
a.b.c.d
a.b.c.d
...

Out
Label
5
5
5
5
...

If the traffic from different CNs have different QoS requirements, HA needs to establish a new LSP from HA to FA for each
class of service. That means that HA must know the number of
CoS of the traffic destined to MN home address. When packets arrive at HA, it needs to classify the packets to identify its
CoS and destination. Then, HA maps them to the corresponding LSP based on the combination of the CoS and destination
address of the packets. Using such mechanism, we can support
differentiated services in MPLS networks [11].
B. Multiple Domains
Multiple domain connectivity needs to be considered in our
scheme as there is a possibility of mobile nodes moving between
different domains. There are some specific requirements on the
border routers of these domains depending on the nature of the
inter-domain connections as described in the following subsections.

MPLS
Network
FA


MPLS
Network

LSR5

LSR6

MN

FA

Router1

LSR5

IP
Network

FA

MN
LSR3

CN1

LSR1

MPLS
Network

LSR3

HA

LSR2

HA

LSR1


CN1

LSR4

MN

MPLS
Network

LSR2
LSR4

MN

Figure 9. Multiple MPLS Domains

MN

Figure 10. An IP Cloud in Between

B.2 An IP Cloud In-between

When there is an IP cloud between the HA domain and the FA
(Figure 10), an IP tunnel is needed to carry the data packets to
the FA. In this case, LSR3 will act as an interchange between the
LSP and the IP tunnel, acting as the FA from the viewpoint of
the HA. Packet is switched from the HA to LSR3 and tunneled
from LSR3 to the FA. Here the hierarchical FA management
scheme can be a solution [13], where every edge router has to
be a hierarchical FA. Slight modification can be made if the FA
is in a MPLS domain. The IP tunnel can be terminated at LSR5.
A LSP will continue the data forwarding task from LSR5 to the
FA. This modification requires LSR5 to be Mobile IP enabled.
In this case, the performance of the proposed scheme become
worse than all previous cases. But in any case, the IP tunnel is
shorten in the proposed scheme. Since switching is faster than
conventional IP forwarding, the transmission delay is improved.
C. Implication of Schemes
We have considered the case where the whole network in
question is a single MPLS domain, multiple MPLS domains and
the case where there are non-MPLS clouds present. Here the key
of different inter domain connectivity is whether the HA packet
processing needs to go up to the IP layer. If it does, then IP
tunneling has to be used to extend the LSP to FA. Our scheme
works in all these possible cases. However, our scheme works
best in the case where the whole network is MPLS capable.
VI. E VALUATION
To evaluate the MPLS and Mobile IP integration scheme performance, we built a testbed and designed a set of experiments
to analyze the scheme.
A. Testbed
The single domain case has been implemented and evaluated
on Linux 2.3.30 software platform. The experimental results reported in this paper are based on measurements taken from the
testbed illustrated in Figure 11. It consists of four PC routers
based on multi-homed 133MHz Pentium PCs hardware. They
are CN, HA, FA and one intermediate LSR between the HA and
FA. CN sends the packet destined to MN to HA and the HA forwards it to the FA through the intermediate LSR between them.
All of them are interconnected using 100Mbps full duplex links.
MN is a 133 MHz Pentium PC. HA and FA runs Mobile IP implementation [14] in user space. The MPLS switching function
is implemented in Linux kernel.

last experiment.
HA Processing Delay with Number of Table Entries
600

137.132.153.117 CN

LSR 137.132.153.57

Ethernet

HA Processing Delay (usec)

FA

HA

D. Roundtrip Delay

Pure Mobile IP
Mobile IP over MPLS
Mobile IP and MPLS Integration
500

137.132.153.114

137.132.153.113

137.132.153.241

137.132.153.249
137.132.153.240

In this experiment, we measure the roundtrip delay. We also

increase the routing and label table size from 5 entries to 8000
entries. We measure the roundtrip delay using ping from CN to
MN. We set the packet size as 1000 bytes. Each data point is
an average of 20 consecutive measurements. From Figure 14,
we can find that the roundtrip delay in Mobile IP scheme increases with the increasing routing table size. In MPLS-Mobile
IP integration scheme, the delay is constant. The reason for this
phenomenon is as explained in the first experiment. In addition
to the HA performance improvement, it also benefits from fast
switching because packet is label switched along the whole path
from CN to FA and back to CN.

400

300

200

Subnet 1
Subnet 2

137.132.153.248

100

MN
137.132.153.250

Figure 11. Testbed

1000

2000

3000
4000
5000
Number of Routing Entries

6000

7000

8000

Figure 12. Processing Delay at HA

B. Processing Delay at HA
During this experiment, we increase the routing and label table size from 5 entries to 8000 entries. The measurements are
plotted in Figure 12. Each point on the graph was obtained by
averaging 50 consecutive measurements. From Figure 12, we
can find that the HA processing delays in Mobile IP and Mobile IP over MPLS schemes increase with the increasing routing
table size. But in the MPLS-Mobile IP integration scheme, the
HA processing delay is almost constant. It is much lower than
the values in Mobile IP and Mobile IP over MPLS schemes. The
lower value is the result of having the entire HA data forwarding process executed in the MPLS layer after MPLS-Mobile IP
integration. So no IP routing table search is executed. Since
label table search is much faster than longest-bit-matching routing table search and IP tunneling needs to search routing table
twice, much processing time is saved and HA performance is
much improved. We also can find that Mobile IP over MPLS
has poorer performance than pure Mobile IP. This is caused by
the additional processing at MPLS layer before the packet goes
up to the IP layer.

VII. C ONCLUSIONS
In this paper, We provided the signaling and control mechanisms to integrate Mobile IP and MPLS. This integration makes
IP-in-IP tunneling in the data forwarding process unnecessary.
Instead we use MPLS to switch the packet. Switching is much
faster than conventional IP forwarding, the transmission delay
and packet processing overhead is reduced. The whole forwarding process is done at the MPLS layer and HA doesnt need to
go up to the IP layer to do the IP tunneling. So the scalability
of Mobile IP is much improved. In addition, since label header
is much smaller than IP header, the traffic overhead from HA to
FA is also reduced.
This work is an initial step towards integrating the MPLS and
Mobile IP protocols. Other future work includes provisioning of
QoS guarantees and route optimization support for our scheme.
R EFERENCES
[1]

Roundtrip Delay Increase with HA Entries Increasing

Throughput Decrease with HA Entries Increasing

13.2

485

[2]

Pure Mobile IP
MIP and MPLS Integration

480

13

[3]
475

Roundtrip Delay (msec)

Throughtput (KBytes/sec)

12.8

470

465

12.6

[4]

12.4

[5]

460

12.2
455

450

[6]

Pure Mobile IP
MIP and MPLS Integration

1000

2000

3000

4000
5000
HA Table Entries

6000

7000

Figure 13. TCP Performance

8000

12

1000

2000

3000

4000
5000
HA Table Entries

6000

7000

Figure 14. Roundtrip Delay

8000

[7]
[8]
[9]

C. TCP Performance
In this experiment, we study the impact of the number of table
entries on Mobile IP forwarding scalability. we also increase the
routing and label table size from 5 entries to 8000 entries. We
measure TCP throughput using ttcp by downloading data from
CN to MN. Each data point is an average of 5 independent measurements. From Figure 13, we can find that the TCP throughput in Mobile IP scheme drops with the increasing routing table
size. In MPLS-Mobile IP integration scheme, the throughput is
constant. The reason for this phenomenon is as explained in the

[10]
[11]
[12]
[13]
[14]

P.C. Mason, J.M. Cullen, N.C. Lobley,UMTS architectures, Mobile

Communications Towards the Next Millenium and Beyond, IEE Colloquium, 1996, Page(s): 401 -411.
Randy H. Katz, et al., ICEBERG: An Internet-core Network Architecture
for Integrated Communications, IEEE Personal Communications.
Andrew T. Campbell, Javier Gomez, Andras G. Valko, An Overview of
Cellular IP, IEEE Wireless Communications and Networking Conference
(WCNC99), New Orleans, Sept. 1999.
D. OMahony, UMTS: the fusion of fixed and mobile networking, IEEE
Internet Computing, Volume: 2 1, Jan.-Feb. 1998, Page(s): 49 -56.
E.Rosen, A. Viswanathan, R. Callon, et al., Multiprotocol Label Switching Architecture, IETF Internet Draft, Aug. 1999.
L. Andersson, P. Doolan, N. Feldman, Fredette, B. Thomas, et al., Label
Distribution Protocol, IETF Internet Draft, Jun. 1999.
C. Perkins, ed., IPv4 Mobility Support, RFC 2002, Oct. 1996.
H.H.-Y. Tzeng, T. Przygienda, On Fast Address-Lookup Algorithms,
Selected Areas in Communications, IEEE Journal on Volume: 17 6, Jun.
1999, Page(s): 1067 -1082.
L. Andersson, A. Fredette, B. Jamoussi, R. Callon, et al., ConstraintBased LSP Setup using LDP, IETF Internet Draft, Nov. 1998.
D. Awduche, J. Malcolm, J. Agogbua, et al., Requirements for Traffic
Engineering Over MPLS, IETF Internet Draft, Jun. 1999.
I. Andrikopoulos, G. Pavlou, Supporting Differentiated Services in
MPLS Networks, Proc. IWQoS 99, Seventh International Workshop on
QoS, 1999 , Page(s): 207 -215.
Y. Rekhter, Eric Rosen, Carrying Label Information in BGP-4, IETF
Internet Draft, Jan 2000.
Charles E. Perkins, Mobile-IP Local Registration with Hierarchical Foreign Agents, 22 February 1996.
The Mobile IP implementation done at the National University of Singapore is avaiable at http://mip.ee.nus.edu.sg.