Supporting Hierarchy and Heterogeneous Interfaces in
Multi-Hop Wireless Ad Hoc Networks
Josh Broch David A. Maltz David B. Johnson
Computer Science Department
Carnegie Mellon University
Pittsburgh, PA 15213
http://www.monarch.cs.cmu.edu/
Abstract
Much progresshas been made toward solvingthe problem of routing packets
inside an ad hoc network, but there are presently no complete proposals
for connecting ad hoc networks together to form larger networks, or for
integrating them with wired internets. This paper describes a technique that
allowsasingleadhocnetworktospanacrossheterogeneouslinklayers. Using
this technique, we can both integrate ad hoc networks into the hierarchical
Internet and support the migration of mobile nodes from the Internet into
and out of ad hoc networks via Mobile IP. Taken together, these solutions
improve the scalability of flat ad hoc networks by introducing hierarchy,and
they enable all nodes participating in the ad hoc network to be reachable
from anywhere in the world. We have implemented each of the solutions in
a real testbed of 8 nodes using the DynamicSource Routing(DSR) protocol.
Generalizingoursolutions,we describeseveralabstractscenariosandpresent
our ideas for solving them.
1 Introduction
In areas in which there is little or no communication infrastructure,
or the existing infrastructure is expensive or inconvenient to use,
wireless mobile users may still be able to communicate through the
formation of an ad hoc network. In sucha network, eachmobilenode
operates not only as a host but also as a router, forwarding packets
for other mobile nodes in the network that may not be within direct
wireless transmission range of each other. Each node participates in
an adhocrouting protocolthat allows it to discover “multi-hop” paths
throughthenetworktoanyothernode. Theideaofadhocnetworking
is sometimes also called infrastructurelessnetworking[10], sincethe
mobile nodes in the network dynamically establish routing among
themselves to form their own network “on the fly.” Some examples of
the possible uses of ad hoc networking include students using laptop
computers to participate in an interactive lecture, businessassociates
sharing information during a meeting, soldiers relaying information
for situational awareness on the battlefield [7, 12], and emergency
disaster relief personnel coordinating efforts after a hurricane or
earthquake.
Inordertodeployad hocnetworksinscenariossimilartothose just
described, ad hoc network routing protocols will be required to sup-
port different types of network interfaces. For example, two teams
of disaster relief personnel from different organizations may have
different types of network interfaces, but they will still need to com-
municate effectively and efficiently. Although there are numerous
proposals for ad hoc network routing protocols, none of the existing
This work was supported in part by the National Science Foundation (NSF) under
CAREER Award NCR-9502725, by the Air Force Materiel Command (AFMC) under
DARPA contract number F19628-96-C-0061, and by the AT&T Foundation under a
Special Purpose Grant in Science and Engineering. David Maltz was also supported
under an Intel Graduate Fellowship. The views and conclusions contained here are
those of the authorsand shouldnot be interpretedas necessarily representingthe official
policiesor endorsements,either expressor implied,of NSF, AFMC, DARPA, theAT&T
Foundation,Intel, Carnegie Mellon University, or the U.S. Government.
protocols fully address the issues of supporting heterogeneous net-
work interfaces, using heterogeneous interfaces to achieve scalabil-
ity, and interconnecting with the Internet. In this paper, we describe
the initial design of an addressing architecture that solves these prob-
lems. We haveimplemented the architecturein a real ad hocnetwork
testbed [9] using the Dynamic Source Routing protocol (DSR) [3,
5, 1] and Mobile IP [11, 4].
Section2 of this paper provides an overview of the basicDynamic
Source Routing protocol (DSR). Section 3 details our addressing
architecture, while Sections 4, 5, and 6 explain how the addressing
architecturecanbeusedto supportheterogeneousinterfaces, connect
an ad hoc network to the Internet, and provide Mobile IP support
within an ad hoc network, respectively. Section 7 explains three
general problems and our current approach to solving them.
2 Dynamic Source Routing
The Dynamic Source Routing protocol (DSR) [3, 5, 1] works by
discovering and using source routes. That is, the originator of a
packet first learns the complete, ordered sequence of network hops
necessary to reach the destination, and each packet sent carries this
listofhops initsheader. Thekeyadvantage ofa sourceroutingdesign
is that intermediate nodes do not need to maintain up-to-date routing
information in order to route the packetsthat they forward, since the
packetsthemselvesalready contain allof the routing decisions. This
fact, coupled with the on-demand nature of the protocol, eliminates
the need for the periodic route advertisementand neighbordetection
packets present in other protocols [2].
The DSR protocol is composed of two mechanisms: Route
Discovery and Route Maintenance. Route Discovery is the mecha-
nism by which a node S wishing to send a packet to a destination
D obtains a source route to D. To perform a Route Discovery, the
source node S broadcasts a R
OUTE REQUEST packet that is flooded
through the network in a controlled manner and is answered by a
R
OUTE REPLY packet from either the destination node or another
node that knows a route to the destination. To reduce the cost of
Route Discovery, each node maintains a cache of source routes it
has learned or overheard, which it aggressively uses to limit the
frequency and propagation of ROUTE REQUESTs.
When sending or forwarding a packet to some destination D,
Route Maintenance is used to detect if the network topology has
changed such that the route used by this packet has broken. When
a route breaks, the detecting node returns a R
OUTE ERROR packet to
the original sender S of the packet. The sender S can then attempt
to use any other route to D that is already in its route cache, or can
invoke Route Discovery again to find a new route.
3 Addressing Architecture
Amongthe most basicproperties of a networkis themannerin which
the nodes of the network are assigned the addresses by which other
nodes will communicate with them. For this discussion, we define
a node in the ad hoc network to be an entity capable of moving
independently from the other nodes in the network. A group of
computers that always move together, such as a wired network of
componentson a vehicle, can be handledby recursively applyingthe
techniques described in this paper.
In the most general case, each node in an ad hoc network will
be acting as an independent router. This implies that the addressing
scheme inside an ad hoc network should ideally be flat, meaning
that each address serves only as an identifier and does not convey
any information about where one node is topologically located with
respect to any other node. For any type of hierarchical addressing
scheme inside a single ad hoc network to make sense, nodes would
have to be constrained to move together with the other nodes in their
branch of the hierarchy, or the hierarchy of addresses would have to
be continually updated as nodes move. Such movement constraints
would violate the spirit of an ad hoc network as a collection of equal
peers opportunistically using each others’ services to communicate,
and the continual reassignment of addresses could become a very
expensive proposition, depending upon the rate of node movement.
Althougha singlenode mayhave manydifferent physicalnetwork
interfaces, which in a typical IP network would each havea different
IP address, we would like each node in the ad hoc network to have
a single identifier by which it is known to all other nodes in the
network. This allows each node to be recognized by all other nodes
in the ad hoc network as a single entity regardless of which interface
they use to communicate with it. We therefore require that each
node participating in the ad hoc network select a single IP address
from the ones assigned to it and that it use only that address when
participating in the DSR protocol. In keeping with the terminology
usedby Mobile IP, we refer to this addressas a node’s home address.
The selection of a single address is important because if a node
were to use multiple addresses when participating in the DSR pro-
tocol, two source routes which pass through the same nodes in the
same order could contain different sequences of IP addresses. This
reduces the ability of Route Discovery to reuse paths to destinations
that othernodesmay have in their route caches,and greatly increases
the workrequired of Route Maintenanceto purge invalid routesfrom
the caches of nodes in the network.
Since each node is known to other nodes by a single IP address,
some other notation is required to distinguish between the multiple
network interfaces a node might carry. Under our addressing archi-
tecture, each node locally assigns a unique interface index to each
of its network interfaces. In most operating systems, this is already
done; for example, the if
index field in the ifnet structure of
BSD Unix-based networking stacks [13] serves this purpose. With
theexceptionof severalreservedindices, theseindexvaluesare local
toeachnode,and the indexvalueschosenbya node havenomeaning
outside of that node except to represent a unique network interface.
This eliminates the need to globally agree on a mapping between in-
terface indices and interface types and allows nodes to encode extra
information that is locally significant into the index value.
We define a path through the ad hoc network from a source node
N
0
to a destination node N
m
as a source route consisting of a series
of hops N
0
/
i
0
!
N
1
/
i
1
!
N
2
/
i
2
!
:::
!
N
m
.WeuseN
k
/
i
k
to
indicate that node N
k
must transmit the packet out its interface
i
k
in
order to deliver the packet over the next hop to node N
k
+
1
.
G2
G1
G3
Figure 1 Clouds of nodes communicating via short-range radios
and gateway nodes with both short-range and long-range radios.
Each cloud may be multiple network hops in diameter.
The nodes in an ad hoc network can have their home addresses
assigned using many different mechanisms, subject to the basic re-
quirement that the addresses be unique inside the ad hoc network.
If the ad hoc network is guaranteed to never connect to any other
internet, then the addresses are only opaque identifiers and can be
drawn from any uniquenumberingspace. For example,a nodecould
select the lowest MAC address from its network interfaces cards as
an address.
In contrast, when groups of nodes are expected to work together
as an ad hoc network and internetwork with other nodes via an
IP internet, their home addresses can be assigned from a single IP
subnet just as would be done for wired hosts. This does not imply
that any hierarchy exists within a single ad hoc network, but rather
that a single ad hocnetwork is a subnet within the hierarchy of some
IP internet. We explain how a node can migrate from one ad hoc
network to another in Section 6.
Assigning the home addresses from the same legal IP subnet
provides several benefits. First, it facilitates connectivity with the
Internet (Section 5), since the border routers that connect the ad hoc
network to the rest of the Internet can distinguish between IP ad-
dresses which are homed inside the ad hoc network and external
addresses. Second, the border routers can advertise reachability to
the ad hoc subnet on the Internet using the standard Internet routing
protocols since each of the nodes in the subnet has a legal routable
IP address. Third, as discussed in Section 7, it can be used to artifi-
cially limit the size of a single ad hoc network, and thereby increase
scalability by breaking a large ad hoc network into several smaller
ad hoc networks.
4 Handling Heterogeneous Interfaces
Onecommonarchitecture forad hoc networksis depictedin Figure1
where clouds of nodes with one type of wireless network interface
are gathered together with gateway nodes with two or more types of
network interfaces.
Such an architecture is an example of an overlay network [6],
where the dashed lines between square boxes represent a long-range
radio used to connect the clouds of nodes, which in turn use short-
range high-speed radios to communicate among themselves. For
example, in a military setting a company of soldiers might use
short-range radios to communicateamong themselveswhile relaying
through a truck-mounted satellite system to communicate with other
companies. In an office setting, each room might have a basestation
interconnected by a wired network while mobile nodes using short-
range infrared transceivers form a multi-hop cloud of nodes in each
room.
AB
CD
reversereverseforward
route
A/1
Route Request for D
route
A
reverseforward
route
A/1
Route Request for D
route
A
B/1 B/2
C/4 C/4
121 14
forward
route
A/1
Route Request for D
route
A
B/1
B/2
Figure 2 Route Discoveryin a ad hoc network
with heterogeneousnetwork interfaces.
In the most generalcase, nothing can be assumed aboutthe home
address each node uses or the arrangement of nodes into clouds.
Since each cloud may be several transmission hops across, all the
nodes in the network must participate in the ad hoc network routing
protocol in order to communicate even with the other nodes and
gatewaysin theirowncloud. Eventhoughtherearedifferent network
interfacesin useand a hierarchyof cloudsis apparentin Figure 1,the
routing protocol must treat the entire network as a single flat routing
domain since it does not know aprioriwhich cloud a given address
can be found in.
The addressing architecture detailed in Section 3 gives DSR the
ability to treat the overall network as single routing domain since
the use of interface indices allows a source route, and thus a Route
Discovery, to traverse interface types.
Figure 2 shows an example of an ad hoc network with hetero-
geneous network interfaces. Node A is using one type of network
interface (represented by the triangles), nodeC and node D are using
an entirely different type of physical network interface (represented
by the circles), and gatewaynodeB is a multi-homed ad hoc network
nodethatcan route betweenthe twodifferent typesof radio technolo-
gies. As described in the previous section, each node independently
chooses an interface index for its interfaces, so that while B and D
have both chosen index 1 for their circle interfaces, C has chosen
index 4.
The example in Figure 2 shows how a R
OUTE REQUEST for D
originated by A will propagate across the network. As the REQUEST
propagates it will collect both a forward route from A to D and a
reverse route from D to A
1
.WhenA’s ROUTE REQUEST is received
by B, B checks if it is already listed on the source route recorded
in the packet or has already repropagated a copy of this REQUEST.
If neither is true, B adds itself to the listed route and repropagates
the REQUEST out all its interfaces, including the one it was received
on. When B transmits the packet out interface
i
, it lists itself in
the forward route as B/
i
. C receives the request and repeats this
process, so that when the packet is received at D it contains both a
route from A to D and a route from D to A. D returns the discovered
route, A/1
!
B/1
!
C/4
!
D,toA in a ROUTE REPLY packet. D
may return the REPLY to A using a cached route, using the reverse
route accumulatedin the R
EQUEST,or by doingRoute Discoveryand
piggybacking the REPLY on its request for A.
The packet headersin Figure 3 show how the source route would
be used to route a packet from A to D, with the outlined boxes
indicating which hop in the source route is being processed. This
exampledemonstratesthe need for a source route to include both the
home address and interface index of each hop. Otherwise, node B
would not have the information necessary to determine which of its
1
Although each node’s address is shown twice in each packet in Figure 2, in the
actual packet format used, each address appears only once, together with the interface
index for the forward route and the reverse route at each node.
AB
CD
A/1
D
C/4
B/1
A/1
D
C/4
B/1
A/1
D
B/1
C/4
121 14
Figure 3 Thesource route ona packetas it movesthrough an ad hoc
network changing physical interface types from triangle interfaces to
circle interfaces. The outlined boxes indicate which entry in the
source route is used when transmitting the packet at each stage.
interfaces should be used when forwarding the packet. Once this
information is present in each packet, packets can be routed seam-
lessly across heterogeneous network interfaces without any further
additions to the system.
Althoughtheexamplesherehaveusedonly onegatewaypercloud
and only two types of interfaces, there is no limit to the number of
gatewaysinasinglecloud,nortothenumberofinterface types. Since
each R
OUTE REQUEST packet builds up a source route of the path it
has travelled across the network, and since each gateway inserts its
unique home address into each R
OUTE REQUEST it propagates,DSR
RouteDiscovery acrossheterogeneousinterfacesis guaranteed to be
trivially loop-free just as it is acrossa network with homogeneousin-
terfaces. Additionally, the same optimization thatcauses each nodein
a homogeneousnetworkto only repropagatea ROUTE REQUEST once
also works in heterogeneousnetworks, causingROUTE REQUESTS to
flood fill the network in an orderly fashion.
5 Integration with Internet Routing
Another issue that the addressingarchitecture describedin Section3
solves is the problem of connectingan ad hocnetworkto theInternet.
Since routing within the ad hoc network is flat, and routing within the
Internet is hierarchical, it is necessary to provide the illusion to the
outside world that the ad hoc network is simply a normal IP subnet.
Local delivery within the ad hoc “subnet” is accomplishedusing the
DSR protocol (possibly over many hops) while standard IP routing
mechanismsdecide whichpacketsshouldenterand leave the subnet.
Figure 4 depicts how an ad hoc network can be connected to the
Internet. Node G1 is a gateway (border router) between the ad hoc
network and the Internet. Routing on G1’s interface internal to
the ad hoc network is accomplished using DSR, while its interface
connected to the Internet is configured to use normal IP routing
mechanisms.
In order for a node A within the ad hoc network to communicate
with a node D outside of the ad hoc network, A simply initiates
Route Discovery (Section 2) for D.AstheR
OUTE REQUEST from A
targeting D propagates,it is eventually received by the gatewaynode
G1, which consults its routing table. If G1 believes D is reachable
outside the ad hocnetwork, it sends a proxy reply listing itself as the
second-to-last node in the route and D as the last node in the route.
Whengeneratinga proxy reply, the reserved gateway interface index
(253) is used to distinguish this reply from normal R
OUTE REPLYs.
Whennode Asubsequentlyoriginatesa datapacket fornodeD,the
source route on the packet will be A/1
!
B/1
!
C/1
!
G1/253
!
D.
When node G1 receives the packet for D it will notice the reserved
gateway interface index in the source routing header, remove the
source routing header from the packet, and transmit the packet out
its interface to the Internet. This packet will have an IP source
address of A and an IP destination address of D and is identical to a
Route Reply
A,B,C,D
A,B,C,G1,D
Proxy
Route Reply
Route Request
D
A
B
C
F
G
E
G1
D
Internet
Route to D?
A,B
A
A,B,C
Figure 4 AROUTE REQUEST for node D being
answered by D and by the gateway node G1.
packet that A would send to node D if it were attached to a normal
IP subnetinstead of a DSR ad hoc network.
Ifthe targetnodeD isactuallyinsidethe adhocnetwork(Figure 4)
thennode A will receive a ROUTE REPLY from both G1 andD.Since
the REPLY from D will not contain a gateway interface index, A can
prefer the direct route when sending packets to D.
With the mechanism described above, nodes inside the ad hoc
network can discover routes thatallow them to send packetsto nodes
outside the network. Allowing packetsfrom the Internet to be routed
into the ad hoc network merely requires that the gateway (node G1)
be configuredas a standard IP router for the ad hoc network subnet.
For example, referring to Figure 4, if node D, located somewhere
in the Internet, were to transmit a packetdestinedfor node A, normal
IP routing techniques would be applied to get the packet from D
to G1. After examining the packet, G1 would determine that the
packetis destinedfor a node in its subnetand wouldattempt to route
the packet to A using DSR. If G1 does not have a cached source
route for node A, it performs a Route Discovery. Supposing that
it discovers the source route G1/1
!
C/1
!
B/1
!
A, it would then
insert the source route D/253
!
G1/1
!
C/1
!
B/1
!
A into D’s IP
packet and transmit the packetinto the ad hoc network.
The technique describedin this section to connecta single ad hoc
networktothe Internetcanalsobeappliedtoincreasethecontainment
of Route Discovery [8] in a network of heterogeneous interfaces,
even if the network is not connected to any Internet infrastructure.
Containmentis definedas the fraction of nodesin the ad hocnetwork
that do not overhear a particular R
OUTE REQUEST, and this metric
correlates directly with scalability.
Figure 5 shows three different ad hoc clouds, a shaded cloud, a
white cloud, and a striped cloud, each connected to the other clouds
using long-range radios. Suppose node A in the shaded cloud is
performing Route Discovery for node B in the white cloud. Using
the technique described in Section 4, this Route Discovery would
propagate throughout the entire ad hoc network, bothering nodes in
all three clouds. However, if the home addresses are assigned such
that each cloud is a distinct IP subnet, the multi-homed gateways
(G1, G2,andG3) can be configurednot to forward R
OUTE REQUEST
packets into their cloud if the REQUEST targets an address not be-
longing to their subnet. In our example, the R
OUTE REQUEST would
be contained to the three gateways (G1, G2,andG3) and the white
and shaded clouds; it would not needlessly be propagated into the
striped cloud.
Furthermore, each gateway can proxy reply for nodes in their
cloud. If G2 proxy replies for node B, this decreases the latency
0
0
1
1
00
00
11
11
0
0
0
1
1
1
0
0
0
1
1
1
0
0
0
1
1
1
0
0
0
1
1
1
00
00
00
11
11
11
00
00
11
11
00
00
00
11
11
11
000
00
0
00
0
00
0
00
0
00
0
000
111
11
1
11
1
11
1
11
1
11
1
111
G1
A
B
G2
G3
Figure 5 Hierarchical routing in the
absence of wired infrastructure.
of Route Discovery observed by A. When a packet from node A
to node B arrives at G2, G2 will take responsibility for delivering
the packet to B, performing Route Discovery as necessary. This
is extremely advantageous because topological change in the white
cloud is then completely hidden from node A, meaning that A will
not need to perform Route Discovery simply because B is moving
around inside of its cloud.
6 Integration with Mobile IP
The previous section detailed how the flat addressing scheme of an
ad hoc network could be integrated with the hierarchical addressing
used in the Internet to facilitate communication between nodes in the
ad hoc network and nodes anywhere else in the Internet. Suppose,
however, that in addition to an ad hoc network that is connected to
the Internet, there is also a mobile node whose home network is not
the ad hocnetwork. For some period of time, this mobile noderoams
through the area where the ad hoc network is located and during that
time would like to join the ad hoc network and take advantageof the
ad hoc network to access other resourceson the Internet.
One specific example of this scenario could be a construction site
where each vehicle on the site participates in an ad hoc network. A
technicianmightoccasionallytravelto thesitetoservicethevehicles.
While doing so, this technicianwouldlike tojoin the ad hoc network
so that he can use it to access manual pages or other resources at his
home office which is connected to the Internet.
The primary mechanism that we use to support visiting mobile
nodes is Mobile IP. Suppose that node MN in Figure 6 is a mobile
node not homed within the ad hoc network and that node FA is a
gateway between the ad hoc network and the Internet that provides
Mobile IP foreign agent services.
The mobile node (MN) will typically keep its network interface
in promiscuous receive mode and so will know that it has en-
tered a DSR network when it overhears DSR packets like R
OUTE
REQUESTs, ROUTE REPLYs or data packets with DSR source routes
on them. After node MN decides to participate in the ad hoc net-
work, it will transmit a Mobile IP A
GENT SOLICITATION piggybacked
on a ROUTE REQUEST targeting the IP limited broadcast address
(255.255.255.255). This allows the SOLICITATION to propagate over
multiple hops through the ad hoc network, though gatewayswill not
propagateitbetweensubnets. WhenFA receivestheSOLICITATION,it
will reply with an AGENT ADVERTISEMENT, allowing MN to register
itself with this foreign agent and with its home agent as a Mobile IP
mobile node visiting the ad hoc network. Once the registration is
complete,the mobilenode’s home agent will useMobile IP to tunnel
packets destined for mobile node MN to foreign agent FA and FA
will deliver the packets locally to the mobile node using DSR.
B
C
F
FA
E
HA
G
CN
MN
MN
Internet
home network
Need FA
Agent
Advertisement
Solicitation
Agent
Figure 6 A visiting mobile node registering with a
foreign agent (FA) in the ad hoc network.
7 General Problems
The techniques described thus far successfully enable (1) the use
of heterogeneous interfaces, (2) the integration of an ad hoc net-
work into the Internet as a subnet, and (3) the movement of mobile
nodes into and out of an ad hoc network using Mobile IP. This
functionality has been completely implemented and tested in our
physical ad hoc network testbed, which has been in operation since
December 1998 [9].
These techniques improve the scalability of an ad hoc network in
situations where nodes in different ad hoc clouds can only commu-
nicate via the gateways. This enables the gateways to contain Route
Discoveries because they have enough information about the hierar-
chy of subnetsto proxy reply for their cloud and to determine which
R
OUTE REQUESTs can safely be excluded from their cloud. This
section explores the more general cases in which the ad hoc clouds
can directly interact, and presents our current ideas for solving the
new problems that arise.
7.1 Overlapping Ad Hoc Clouds
Although nodes in an ad hoc network may often be arranged into
clouds containing gateway nodes with multiple interfaces, and these
clouds may have been formed with addresses drawn from the same
subnet, it may frequently be the case that these clouds overlap spa-
tially. As shown in Figure 7, some nodesfrom the shaded, white and
striped clouds are in range of each other via their short-rangeradios.
In this environment, if a shaded node transmits a R
OUTE REQUEST
for a whitenode, the requestwill directly floodthe entire networkvia
the short-range radios. The fact that the multi-homed square nodes
have beenconfiguredto proxy reply on behalf of their subnetclouds
will not allow them to contain the Route Discovery as in Section 5.
The spread of a Route Discovery across the entire network is a
concern because the number of overhead packets required by the
routing protocol typically increases with the number of nodes in the
routingdomain. We are just beginningto studythescalingproperties
of DSR with respect to the number of nodes in the routing domain,
though initial simulations show that DSR performs well with 25, 50,
and 100 nodes. While we have yet to simulate larger networks, we
believe the maximum practical size of a routing domain that DSR
canefficientlyhandle,given the optimizationswe haveexperimented
with so far, will be on the order of 500 nodes.
In order to contain Route Discovery, we need a mechanism to
restrict a R
OUTE REQUEST packet originated by a node in one cloud
from being propagated by nodes homed in a different cloud. This
will keepthenodeslogicallyseparateeventhoughthey arephysically
co-located. We assume that each cloud is a separate IP subnet, and
000
00
0
00
0
00
0
00
0
00
0
111
11
1
11
1
11
1
11
1
11
1
00
00
11
11
00
00
00
11
11
11
00
00
11
11
00
00
00
11
11
11
00
00
11
11
00
00
00
11
11
11
0
0
1
1
00
00
11
11
0
0
1
1
0
0
1
1
G1
G2
G3
Figure 7 An ad hoc network where the clouds of nodes overlap and
are in wireless transmission range of each other.
that each node is configured with the netmask of the subnet it is a
part of. This allows each node to determine whether or not another
node’s address is inside its subnet, and hence its cloud. Gateway
nodes must be configured with a netmask that identifies not only
their own subnet, but also the subnets of their peers, as belonging to
their same cloud.
We require that nodes only repropagatea ROUTE REQUEST packet
if the R
EQUEST was last propagated by a member of the same cloud
(i.e., the last address in the source route carried by the REQUEST is
from the same subnet). This rule results in a REQUEST originated
by a shaded node only being repropagated by other shaded nodes,
and thereby prevents the Route Discovery from spreading directly
betweenclouds. Thisfiltering ruleappliesonlyto forwarding ROUTE
REQUESTS and not to forwarding packets, so that if a source route is
somehow discovered that crosses directly between clouds, packets
may flow along it.
We must provide an exception to the filtering rule, however, to
handle cases in which a node legitimately intends to have its ROUTE
REQUEST propagated by nodes outside its subnet. A node invokes
the exception by setting the “I” bit in the ROUTE REQUEST.A
node receiving a packet with the “I” bit set will ignore the filtering
rule, add itself to the recorded source route, and clear the “I” bit
before repropagatingthe request. The “I” bit is clearedso that future
propagations will obey the filtering rule.
The “I” bit is used to introduce the R
EQUEST to a new cloud, as it
permits nodes from the new cloud to repropagate the request once.
Because the source route then ends with an address from the new
cloud,other nodes in the newcloud will repropagateit. For example,
when a gateway needs to forward a ROUTE REQUEST into a cloud to
which its home address does not belong,it sets the “I” bit in order to
introduce the R
EQUEST to the nodes in the cloud.
7.2 Wandering Nodes
A different scenario is depicted in Figure 8. In this figure, the white
ad hoc network cloud and the shaded cloud do not overlap spatially,
but one node from the shaded cloud has wandered into the white
cloud, becoming partitioned from the rest of the nodes in its home
cloud. This problem is exactly the problem described in Section 6
and is solved using Mobile IP; gateway G2 acts as a foreign agent
and gateway G1 acts as a home agent for the shaded node that is
visiting the white cloud. This allows the shaded node to continue
communication just as if it were still connected to its home cloud.
Because the shaded node is completely surrounded by white
nodes, it must set the “I” bit on R
OUTE REQUEST packets that it
originates which contain Mobile IP AGENT SOLICITATIONs for a for-
eign agent. This results in the SOLICITATIONs spreading through the
clouds neighboring the shaded node and finding a nearby foreign
agent.
