Internet Connectivity for Multi-Homed

Proactive Ad Hoc Networks

Paal E. Engelstad

Andreas Tnnesen

Andreas Hafslund

Geir Egeland

Telenor R&D
Fornebu, Norway

Thales Communications
Oslo, Norway

University of Oslo (UniK)

Kjeller, Norway

Telenor R&D
Fornebu, Norway

paal.engelstad@telenor.com

andreto@unik.no

andreha@unik.no

geir.egeland@telenor.com

AbstractA prerequisite for a widespread and successful

deployment of proactive ad-hoc networking technology is its
ability to provide easy access to the Internet. Normally, proactive
routing protocols provide routing messages that establish default
routes to ensure connectivity for outgoing IPv4 packets destined
for the Internet. However, mechanisms to ensure connectivity for
incoming IPv4 packets from the Internet are yet poorly
documented in published material. Possible solutions include
implementing a modified Mobile IPv4 Foreign Agent (MIP-FA)
or Network Address Translation (NAT) on each Internet
Gateway node in the ad hoc network. In this paper we discuss
different strategies for providing Internet access to proactive ad
hoc networks. We also describe problems experienced in our lab
test-bed with default routes under the condition of site multihoming. Based on this experience, we propose working solutions
for Internet access from proactive ad hoc networks.
Keywords - manet;Internet; proactive;multi-homing; NAT,

I.

INTRODUCTION

IP-based applications, such as web browsing, e-mail, telnet

and ftp, mainly communicate with servers or peers over the
Internet. A mobile ad-hoc network (MANET [1]) has no fixed
infrastructure, and services on the Internet might not be
available in such networks. A likely scenario is that nodes on
an ad-hoc network in some cases also want to connect to nodes
on the Internet, using services available there. For a widespread
and successful deployment of MANETs, the ability to provide
easy access to the Internet is therefore a prerequisite.
A common approach is to let a MANET node with Internet
access operate as Internet gateways and provide Internet access
to other nodes in the MANET. There can naturally be several
MANET nodes operating as gateways on the MANET at the
same time.
In this paper we address the lack of a good mechanism for
IPv4 Internet access in a proactive MANETs, i.e. a MANET
that is routed with a table-driven routing protocol, such as the
Optimized Link State Routing (OLSR [2]) protocol or the
Topology Dissemination Based on Reverse-Path Forwarding
(TBRPF [3]) routing protocol.
Proactive routing protocols normally allow gateways to use
special routing messages to set up default routes on the
MANET. IP packets that do not have an IPv4 destination
address known locally on the MANET are forwarded along the
default route out of the MANET through the gateway.
Normally, the destined external host on the Internet will
send return traffic to the source IP address of the outgoing

packet. Thus, for IPv6, a MANET node configures an address

under a global prefix managed by one of the gateways [4] and
uses this address as source IP address when communicating
with external hosts on the Internet. Return traffic from the
external nodes on the Internet is therefore routed back to the
gateway, which in turn can forward the packets to the MANET
node.
However, for IPv4 - which is the focus in this paper - there
is a great scarcity of global IPv4 addresses. Thus, the gateway
may be equipped with a very limited number of external IPv4
addresses. To allow different MANET node to share an address
for external communication, the gateway may implement a
Network Address Translator (NAT) [5] or a Mobile IP Foreign
Agent (MIP-FA) [6]. Both solutions have been proposed and
have been implemented for MANET gateways [7-9].
Since MANETs normally are without infrastructure and
have limited capabilities for internal coordination, it is difficult
to control which of the MANET nodes operate as gateways and
which mechanism (e.g. NAT or MIP-FA) each gateway
implements. Thus, there might be both NAT-based and MIPFA based gateways present on the MANET simultaneously.
This issue has not been studied in previous work that proposes
the use of MIP-FA gateways for proactive MANETs [7-8].
This paper explains problems with Internet access for a
multi-homed MANET, i.e. a MANET where more than one
MANET node operate as gateways, and where gateways may
use different gateway technologies. Much of the problem is
caused by the use of default routes that proactive protocols use
to get packets routed out of the MANET.
Based on experience from our test-bed, we propose
solutions for multi-homed MANETs, which work well both for
NAT-based and MIP-FA-based gateways. Although our main
focus is on the OLSR routing protocol, the analyses and
proposed solutions are also applicable to other proactive
routing protocols, such as TBRPF.
Section II presents in depth solutions for Internet
connectivity on proactive MANETs. In Section III we use a
test-bed experiment to show deficiencies of using default
routes in a multi-homed network with NAT-based gateways.
Instead, we propose a new working solution for Internet
connectivity. The relevance of our experimental results to MIPFA based gateways is discussed in Section IV. Here, we
propose a working solution also for this scenario. Section V
shows how to achieve dynamic change of NAT-based
gateways, and Section VI presents experimental results on
performance constraints with mobility and Internet access.
Conclusions are drawn in Section VII.

II.

BACKGROUND

A. Proactive (table-driven) routing protocols

With proactive protocols, the network is periodically
flooded with route information, so that the routing tables
contain complete information of routes to all nodes present on
the MANET. The most widely studied and popular proposals
include the OLSR and the TBRPF protocols. Both protocols
introduce optimizations of the classical link state algorithm
tailored to the requirements of a mobile wireless LAN. Both
provide shortest path routes in terms of number of hops.
OLSR uses multipoint relays (MPRs), which are selected
nodes that forward broadcast messages during the flooding
process. Thus, the message overhead of the flooding process is
reduced substantially compared to a classical link state
algorithm. Furthermore, the number of control messages
flooded in the network is minimized, since it is only nodes
elected as MPRs that generate link state information. An MPR
node may also choose to report only links between itself and its
MPR selectors, allowing partial link state information to be
distributed in the network. The protocol is particularly suitable
for large and dense networks as the technique of MPRs works
well in this context.
TBRPF is based on source trees and reverse path
forwarding. Each node running TBRPF computes a source tree,
based on partial topology information stored in its topology
table, using a modification of Dijkstra's algorithm. The tree
provides paths to all reachable nodes on the MANET. To
minimize overhead, each node reports only a part of its source
tree to neighbors. TBRPF uses a combination of periodic and
differential updates to keep all neighbors informed of the
reported part of its source tree. Each node also has the option to
report additional topology information (up to the full topology),
to provide improved robustness in highly mobile networks.
B. Acquiring Internet connectivity for outgoing traffic
Both OLSR and TBPRF allow a MANET router to
advertise prefixes that are topologically correct for the external
networks to which they are connected. Network prefixes are
advertised in Host and Network Association (HNA) messages
for OLSR or in Network Prefix Association (NPA) messages
for TBPRF. The message binds a set of network prefixes to the
IP address (OLSR) or Router ID (TBPRF) of the node attached
to the external networks. Each message contains one or many
network prefixes, each specified by a netmask (OLSR) or a
prefix length (TBPRF).
The messages are transmitted periodically and the
information expires after a specified time. Each node also
maintains information specifying which nodes may act as
gateways to associated hosts and networks by binding a
gateway address to the prefix of the external network. Upon
reception of a message, a node updates the routing table with
the prefix information contained in the message, before
flooding it further throughout the MANET.
Hence, both OLSR and TBPRF use default routes to
announce reachability to the Internet. A MANET node that has
Internet access over an external network to which it is
connected, operates as a gateway and advertises Internet

connectivity as a 0.0.0.0/0 default route. All packets destined

for addresses without a route on the local MANET, will be
routed out along the default route to the gateway and forwarded
further onto the Internet.
C. Acquiring Internet connectivity for incoming traffic
In addition to using default routes for outgoing packets, a
mechanism is required to ensure that return traffic from the
Internet gets routed back to the MANET.
If the gateway implements a Mobile IPv4 Foreign Agent
(MIP-FA) and if the MANET node runs Mobile IPv4, the
MANET node may register the care-of-address of the gateway
with the Home Agent (HA) [6]. When contacting an external
host on the Internet, it uses its home address as source IP
address. The return traffic is therefore routed to the HA, which
encapsulates the packets and tunnels them to the care-ofaddress of the MIP-FA gateway. The gateway can then inject
the return traffic into the proactive MANET.
A gateway that on the other hand implements Network
Address Translation (NAT) will translate the source IP address
of outgoing packets from the MANET node. It replaces the
source IP address with an address of the NAT gateway, which
is routable on the external network. Hence, an external host
will return packets using the IP address of the NAT-gateway as
destination IP address. The gateway can then replace the
destination IP address with the IP address of the MANET node,
and inject the return traffic into the MANET.
A drawback with a MIP-FA solution is that it requires
changes to the Mobile IP implementation on both the Mobile
Node (MN) side (i.e. on the source node requiring Internet
access) and on the Foreign Agent (FA) side (i.e. on the
gateway). Since the MN and the FA are no longer on-link, both
sides will have to deal with Agent Solicitations and Agent
Advertisements in a different way; TTL values and IP
destination addresses must be set differently; ARP must be
used differently and MAC-addresses are no longer relevant for
communication between MN and FA. The solution differs so
much from regular Mobile IP operation, that a special purpose
"MIPv4-for-MANET" code is probably required (i.e. it is not
beneficial to reuse the regular MIPv4 code).
Moreover, independent co-existing implementations of both
the MANET routing protocol and the "MIPv4-for-MANET"
code are not trivially managed, since both implementations will
make unsynchronized modifications to the routing table.
Another drawback that limits the applicability of a MIP-FA
based solution is that it requires that the care-of-address of the
gateway and the home IP address of MN are globally routable.
However, since the IPv4 address space is a scarce resource,
nodes that require Internet access might only be able to acquire
private IPv4 addresses.
A NAT-based mechanism, on the other hand, appears as an
alternate solution. The NAT functionality may be in the form
of Basic NAT, however NAPT (i.e. NAT with port translation)
is a more applicable solution, since many MANET gateways
might only be able to acquire a single IP-address on the
external network to which they are connected [5]. NATs allow
other MANET nodes to use private addresses. NAT-devices

can be nested, and this solution will work even when the
MANET gateway acquires a private IP address from the
external network.
III.

INTERNET CONNECTIVITY USING NATS

A. UDP traffic and default routes

We tested site multi-homing in a proactive MANET with
two NAT-based gateways present. The source node (SN)
communicated with an external host (XH), and could reach
gateways GW1 or GW2 via the Intermediate Node IN1 and
IN2, respectively. The test configuration is illustrated in Figure
1.

is recalculated, and the SN uses IN2 as the default route to the

Internet, which in turn uses GW2 as next hops of the default
route.
The experimental results are shown in Figure 2. The UDP
traffic alternates between GW1 and GW2. However, there is an
intermediate period after the SN has switched between IN1 and
IN2 where all traffic is dropped, before the packets again get
correctly routed out through one of the gateways. The figure
shows that with the default hello interval of OLSR of 2
seconds, it may take up to 6 seconds (i.e. 3 hello packets) after
the SN has moved from IN1 to IN2 to discover that the route to
IN1 is no longer valid.
250

200

Rate (kbps)

150

100

50

udp,S->D via GW1

udp,S->D via GW2
0
0

20

40

60

80

100

120

140

Time (sec)

Figure 1. Test-bed implementation with two NAT-based gateways (GW1

and GW2). The Source Node (SN) is communicating with an external host
(XH) over the Internet. All external communication passes through an
Intermediate Node (IN1 or IN2).

The OLSR_UniK implementation for Linux [10] was used

as routing modules on all MANET nodes (i.e. SN, IN1, IN2,
GW1 and GW2), and WLAN 802.11b was used for the
wireless communication. All nodes were located in the same
room (5-by-5 meters), and the IP-tables feature of Linux was
used to emulate that the SN was not in direct radio-range with
the gateways.
We added mobility to the network by allowing the SN to
alternate between being within radio range of IN1 and being
within radio range of IN2, on 20 seconds time intervals. That
is, the SN was first within radio range of IN1 and outside radio
range of IN2 for 20 seconds. Then, after 20 seconds, the SN got
within radio range of IN2 and outside radio range of IN1, and
so forth. Before we started measurements, the network was
granted a transient period of some minutes to ensure
convergence of the routing system with respect to the nonmobile nodes. After the transient period, the SN started sending
UDP packets of 512 bytes at a transmission rate of 204 kbps
destined for the XH.
The HNA messages from GW1 and GW2 form default
routes to the Internet on a shortest-hop basis. Hence, while
connected to IN1, the SN uses IN1 as the default route to the
Internet, which in turn uses GW1 as next hops of the default
route. However, as soon as it connects to IN2, the default route

Figure 2. Test-bed experimental results of UDP traffic, using default routes

of the original OLSR implementation. The UDP packets were sent from SN to
XH, and SN hopped between IN1 and IN2 on a 20 second interval.

Thus, the SN may use maximum 6 seconds before it

recalculates the HNA route for the gateway and establishes IN2
as the next hop for the outgoing route. Apart from the fact that
there is a required time for the routing to converge, everything
works fine with UDP traffic.
B. TCP traffic and default routes
We did a similar test. This time, however, the SN
established a TCP session with the XH. In this case, we
experienced as expected that a TCP session would always
break as the default route was shifted to a different gateway.
This experimental result is illustrated in Figure 3.
The reason the TCP-connection breaks can be explained as
follows: The first packets were all routed to the NAT module
of one of the gateways, say GW1, corresponding to the
intermediate node, IN1, that the SN was initially connected to.
The source IP address of all these packets were translated by
the NAT module of the gateway.
However, when the SN eventually established connectivity
with IN2 and discovered that connectivity with IN1 was lost,
TCP packets were forwarded via GW2. When the outgoing
TCP packets passed out through the NAT-module of GW2, the
module naturally translated the source IP address of the TCP
packets to a different address than the one used by the NATmodule at GW1. The packets were not recognized when they
finally arrived at the XH, since TCP uses also the source
address to identify the connection. Upon reception of the first

"misrouted" packet, the XH immediately returned a TCPRESET message to the unknown source address (i.e. it was
routed back to the SN through the NAT module of GW2), and
the TCP session broke as shown in the figure.

We repeated the same experiment with the tunneling

solution. The experimental results are shown in Figure 5.
2500

2000

Rate (kbps)

1500

1000

500

tcp S-> D, tunneled via NAT-GW2

tcp S-> D, tunneled via NAT-GW1
0
0

20

40

60

80

100

120

140

Time (sec)

Figure 3. A TCP session may break when there are two NAT-based
gateways and when default routes are being used.

C. Working solution using explicit tunneling

To avoid that TCP sessions break, we propose to use
explicit tunneling to one of the gateways instead of using
default routes (Figure 4).

Figure 5. Test-bed experimental results with explicit tunneling to a NATbased gateway. (The return traffic in terms of TCP ACK packets is not shown,
since the amount of this traffic is small as indicated in Figure 3).

Figure 5 shows that with explicit tunneling all packets are

consistently routed throughout the same gateway even if the
SN is mobile, and hence the TCP session does not break.
The figure also shows that the throughput was
approximately 30% lower when packets are routed though the
gateway that is one hop further away from the SN. This is as
expected, since the packets have to be transmitted by an
additional node. (Further performance results from our test-bed
are detailed in a follow-on paper [13])
IV.

INTERNET CONNECTIVITY USING MOBILE IP

A. Problems with different coexisting gateway technologies

Benzaid et al. [7-8] propose a solution for OLSR with MIPFA based gateways. For outgoing connectivity, the solution
relies on the default routes established by the HNA routing
messages.

Figure 4. Tunneling of packets for external host via gateway, e.g. using IPin-IP encapsulation.

The SN knows that a host is external if it cannot find the

destination IP address in the routing table. Instead of using the
default route, the SN simply tunnels the IP packets e.g. by IPin-IP encapsulation [11] or minimal IP encapsulation [12], to
the IP address of the gateway, which is found in the Originator
ID field of HNA messages. No changes to the OLSR
specification are required. However, gateways must be able to
accept and decapsulate tunneled packets.
Before sending the first packet to the XH, the SN chooses
an appropriate gateway (e.g. the gateway closest to the SN).
The SN consequently uses the same gateway for subsequent
packets belonging to the same communication session, as long
as it has a route to the gateway.

A problem with the use of default routes is the same as

before: the SN cannot control through which gateway an
outbound packet is routed. As demonstrated by our test-bed
experiments presented above, the default route may change as
the SN moves around on the MANET, and packets may be
routed out through different gateways correspondingly.
The use of Mobile IP easily leads to triangle routing. In a
MANET this means that outbound packets may go out through
one gateway (e.g. the closest one) while return packets enters
the MANET via the MIP-FA gateway to which the MANET
node is registered. However, it is likely that this gateway is
behind a firewall. Most firewalls today are stateful; the first
packet going out of the access network sets soft-state in the
firewall and return packets are temporarily allowed to enter the
access network. Thus, if the first outbound packet (e.g. TCPSYN) does not exit the MIP-FA gateway, the return packet
(e.g. TCP-SYN+ACK) will be stopped by the firewall.
Furthermore, in a multi-homing scenario, some gateways
may be NAT-based, others MIP-FA based, and others based on

other technical solutions. This generates a problem for the

solution proposed by Benzaid et al. [7-8], because one gateway
technology may undermine correct functionality of another
gateway technology.
For example, if there are one MIP-FA based gateway and
one NAT-based gateway present on the MANET, outgoing
traffic from a MIP-enabled SN may easily be mistakenly routed
along a default route out through the NAT-based gateway. The
NAT will translate the source IP address and forward the
packet to the external node. Hence, the IP packet will not be
recognized and the communication session (e.g. TCP session)
is likely to break, similar to what we demonstrated above (e.g.
in Figure 3).

gateways instead of using default routes, also for the MIP-FA

based solution (Figure 6).
If the SN uses MIPv4 Reverse tunneling, the explicit
tunneling solution coincides with the Encapsulating Delivery
Style of Reverse Tunneling [15] (Figure 6).

Alternatively, the NAT may drop the packet if it only

accepts tunneled packets according to the explicit tunneling
solution proposed above. In either case, the TCP session will
eventually break.
B. Problems with ingress filtering
Ingress filtering [14] of the outgoing traffic is becoming
more and more common on routers in access networks. Ingress
filtering means that a router will not accept on its ingress
interface packets with a source IP address that is not
topologically correct for that interface. The motivation is to
prevent IP address spoofing.
It is quite probable that many MANET gateways will desire
to implement ingress filtering. Thus, if the outgoing packet
ends up at another gateway than the MIP-FA based gateway
that the SN is registered with, the gateway may drop the packet
to prevent a possible denial-of-service attack.
Ingress filtering might also be implemented on routers
further back in the access network to which the MANET
gateway is connected. Since the SN uses its own home IP
address as source address of outgoing packets, and since this
address would normally not be probably topologically correct
on the access network, SN's packets are easily filtered away.
Ingress filtering is another issue that the MIP-FA solution by
Benzaid et al. [7-8] did not cover.
MIPv4 Reverse Tunneling [15] can be used as an antidote
against this. The mobile node requests reverse tunneling
service from the FA when it registers with it. The FA tunnels
the mobile node's traffic back to the HA, using a topologically
correct IP address as source address of the encapsulating IP
header. The HA will decapsulate received packets and forward
them to the XH.
It is only the MIP-FA-based gateway with which the SN is
registered that will offer reverse tunneling for the SN, and the
HA will not accept packets tunneled from FAs that the SN has
not registered with. As a consequence, it is utmost important
that all outgoing traffic is sent over the MIP-FA based gateway
with which the SN is registered.
C. Working solution using explicit tunneling
To avoid that the TCP sessions break, that packets are
dropped by the gateway, or that packets are subject to ingress
filtering, we propose to use explicit tunneling to one of the

Figure 6. MIP-FA based solution with explicit tunneling to the gateway. The
figure shows the case where SN uses MIPv4 Reverse Tunneling between FA
and HA.

V.

INTERNET CONNECTIVITY USING MOBILE IP OVER

NAT-BASED GATEWAYS

A. Problem with NATs and dynamic change of gateways

A SN can easily change between two MIP-FA-based
gateways without breaking the communication session. The SN
simply registers the new MIP-FA with its own HA. With a
NAT-based gateway, on the contrary, the session of a SN is
bound to the NAT that the session passes through. As the SN
moves around on the MANET, it may move close to another
gateway. In this situation, the SN might desire to change
gateway dynamically without breaking the communication
session.
B. Working solution using NAT traversal
Mobile IP with NAT traversal [16] can be used to enable
dynamic change of NAT-based gateways. The NAT traversal is
based on IP-in-UDP-in-IP encapsulation, using the Mobile IP
Home Agent UDP port 434 for encapsulated data traffic.
The SN must implement a MIPv4 client and uses an
extension in its Registration Request to indicate that it is able to
use Mobile IP UDP tunneling instead of standard Mobile IP
tunneling. If the HA sees that the Registration Request seems
to have passed through a NAT, it sends a successful
registration reply. The home agent will then use MIP UDP
tunneling to the SN, using the same UDP ports and IP
addresses that appeared in the registration request, but in
reverse order. The SN use MIP UDP to tunnel packets back to
the SN, using the same ports and IP addresses as in the original
registration request message. The source port may vary
between new registrations (e.g. when the SN changes to a new
NAT-based gateway), but remains the same for all tunneled
data and for re-registrations. In addition, the SN may
periodically send polling messages to the HA to keep the soft
state mapping in the NAT valid, which otherwise normally
would time out within a couple of minutes.

Since the MANET can be multi-homed, this solution

requires an additional IP-in-IP tunneling from the SN to the
gateway, to ensure that the source IP address - and possibly
also the UDP source port number - are translated consistently
by the same NAT-based gateway, so that the traffic will be
recognized by the HA (Figure 7).

discovered through a new extension to the HNA messages) and

on how far away from the SN it is located.
When the SN wants to send a packet that is destined for an
IP address not present in the routing table, it tunnels the packet
to the selected gateway. All packets belonging to the same
session should normally be tunneled to the same gateway.
We also proposed a solution where source nodes may use
Mobile IP with NAT-traversal to change NAT-based gateways
dynamically. The solution requires 28-bytes worth of extra
overhead in each packet, but as shown in Section III.C. this
can be a small price to pay compared to the bandwidth benefits
of getting traffic routed through the closest gateway.

Figure 7. NAT-based solution with Mobile IP NAT traversal.

A clear advantage of this approach over using the regular

MIP-FA solution is that the SN can rely on existing MIPv4code (enabled for NAT traversal) that it also uses on the
Internet. The SN is not required to implement special "MIPv4for-MANET" code. The solution allows for dynamic change of
gateways between NAT-based gateways and MIP-FA-based
gateways. (However, an additional special "MIPv4-forMANET" code is probably required if the SN shall also be able
to use MIP-FA based gateways.)
A disadvantage is that the solution requires an additional IP
and UDP header's worth of tunneling overhead (i.e. 28 bytes)
between the SN and the gateway. However, this can be a small
price to pay compared to the bandwidth penalty of not routing
packets out through the closest gateway, as was shown in
Figure 5 of Section III.C.
There are also ongoing efforts using header compression
methods to reduce the tunneling overhead of MIP NAT
traversal [17].
VI.

CONCLUDING REMARKS

Solutions for Internet connectivity must be designed for the

possibility that there might be multiple gateways present on the
MANET, and the gateways may implement different gateway
solutions, including NAT and MIP-FA.
In this paper we have shown that these solutions do not
work well in the presence of default routes used by OLSR and
TBRPF. With default routes, outgoing traffic might be directed
over different gateways in a non-deterministic way, depending
on dynamics and mobility in the network. The gateway
solutions, however, require that packets be routed consistently
over a specific gateway.
We have proposed to replace default routes with explicit
tunneling between the SN and the gateway. This requires that
the SN explicitly discovers available gateways, e.g. by means
of the HNA messages of OLSR. Ideally, the SN would select a
gateway based on its capabilities (which might, for example, be

A drawback of the tunneling of outgoing packets is that it

requires additional overhead in each packet, e.g. IP-in-IP
encapsulation would require an additional 20 bytes of overhead
in every outgoing packet. The fraction of additional overhead
will be relatively high for certain types of traffic that require
short packages, such as Voice-over-IP. To save overhead, one
may use minimal IP encapsulation [2] or header compression
mechanisms [16, 18].
The analyses in this paper are also relevant for the
provision of Internet connectivity in IPv6-based MANETs [4].
The use of default routes might be less damaging for IPv6,
since NATs are not anticipated used here. (As we demonstrated
above, only one single packet that was misrouted through a
NAT was enough to close down the entire TCP session.)
However, the problems with ingress filtering and with stateful
firewalls, as described in Section V, are still present in an IPv6
setting.
For IPv6 one may replace IP-in-IP tunneling with a routing
header to source-route outbound packets via the selected
gateway [4]. To reduce the overhead even more, one may omit
tunneling by using source address based default routing [19].
The latter would not be possible with an IPv4-based MANET,
since the source IPv4 address (i.e. either the home address with
MIP-FA based gateway or a private or IPv4 link local address
with a NAT-based gateway) would not correspond to a prefix
of the gateway.
The source code of the NAT-based gateway solution with
explicit tunneling proposed in this paper can be downloaded
from [10].
[1]
[2]

[3]

[4]

[5]

MANET Working Group of the Internet Engineering Task Force (IETF),

homepage, http://www.ietf.org/html.charters/manet-charter.html.
Clausen, T. (ed.) and Jacquet, P. (ed.), "Optimized Link State Routing
Protocol", RFC 3626, Internet Engineering Task Force (IETF), October
2003.
Ogier, R., Templin, F. and Lewis, M., "Topology Dissemination Based
on Reverse-Path Forwarding (TBRPF)", IETF Internet draft, draft-ietfmanet-tbrpf-11.txt, October 2003 (Work in Progress).
Wakikawa, R., Malinen, J., Perkins, C., Nilsson, A., Tuominen, A.,
Global Connectivity for IPv6 Mobile Ad Hoc Networks , IETF
Internet Draft, draft-wakikawa-manet-globalv6-03.txt, October 2003
(Work in Progress).
Srisuresh, P. and Holdrege, M., "IP Network Address Translator (NAT)
Terminology and Considerations", RFC 2663, Internet Engineering Task
Force (IETF), August 1999.

[6]
[7]

[8]

[9]

[10]

[11]
[12]
[13]

[14]

[15]
[16]

[17]

[18]

[19]

Perkins C. (ed.), "IP Mobility Support for IPv4", RFC 3344, Internet
Engineering Task Force (IETF), August 2002.
Benzaid, M., Minet, P., Al Agha, K., Adjih, C., and Allard, G.
"Integration of Mobile-IP and OLSR for a Universal Mobility", To
appear in Wireless Networks journal (Winet), Special Issue on Ad-hoc
Networking.
http://menetou.inria.fr/~benzaid/WINET2003BENZAID.ps.gz
Benzaid, M., Minet, P., and Al Agha, K., "A framework for integrating
Mobile IP and OLSR ad_hoc networking for future wireless mobile
systems", Med-Hoc-Net'02, Sardinia, Italy, September 2002.
Engelstad, P. and Egeland, G., "NAT-based Internet Connectivity for On
Demand MANETs", Proceedings of 1st Wireless On-Demand
Networking Symposium 2004 (WONS 2004), Springer-Verlag LNCS
Series, 2004. (http://www.unik.no/~paalee/PhD.htm).
Tnnesen, A., Implementation of the OLSR specification as specified in
RFC 3626 [2]. Source code can be downloaded from
http://www.olsr.org/
Perkins, C.E., "IP Encapsulation within IP", RFC 2003, Internet
Engineering Task Force (IETF), October 1996.
Perkins, C.E., "Minimal Encapsulation within IP", RFC 2004, Internet
Engineering Task Force (IETF), October 1996.
Hafslund, A., Landmark, L., Engelstad, P., Li, F., "Testing and
Analyzing TCP Performance in a Wireless-Wired Mobile Ad Hoc Test
Bed", To be published. (See http://www.unik.no/~paalee/PhD.htm)
Ferguson, P. and Senie, S., Ingress Address Filtering: Defeating Denial
of Service Attacks which employs IP Source Address Spoofing. RFC
2827, Internet Engineering Task Force (IETF), May 2000.
Montenegro, G. (ed.), Reverse Tunneling for Mobile IP, revised. RFC
3024, Internet Engineering Task Force (IETF), January 2001.
Levkowetz, H. and Vaarala, S., Mobile IP Traversal of Network
Address Translation (NAT) Devices. RFC 3519, Internet Engineering
Task Force (IETF), April 2003.
Vaarala, S., Nuopponen, A., Adrangi, F., Optimized Mobile IPv4 UDP
Encapsulation , IETF Internet Draft, draft-vaarala-mipv4-optudp-00.txt,
January 2004 (Work in Progress).
ROHC (Robust Header Compression) Working Group of the Internet
Engineering
Task
Force
(IETF),
homepage,
http://www.ietf.org/html.charters/roch-charter.html.
Huitema, C., Draves, R., "Host-Centric IPv6 Multihoming" IETF
Internet Draft, draft-huitema-multi6-hosts-03.txt, January 2004 (Work in
Progress).