c
1997 Kluwer Academic Publishers. Printed in the Netherlands.
Associativity-Based Routing for Ad-Hoc Mobile Networks
CHAI-KEONG TOH
University of Cambridge, Computer Laboratory, Cambridge CB2 3QG, United Kingdom
Abstract. This paper presents a new, simple and bandwidth-efficient distributed routing protocol to support mobile computing in a conference size ad-hoc mobile network environment. Unlike the conventional approaches such as link-state and distance-vector distributed routing algorithms, our protocol does not attempt to consistently maintain routing information in every node. In an ad-hoc mobile network where mobile hosts (MHs) are acting as routers and where routes are made inconsistent by MHs’ movement, we employ an associativity-based routing scheme where a route is selected based on nodes having associativity states that imply periods of stability. In this manner, the routes selected are likely to be long-lived and hence there is no need to restart frequently, resulting in higher attainable throughput. Route requests are broadcast on a per need basis. The association property also allows the integration of ad-hoc routing into a BS-oriented Wireless LAN (WLAN) environment, providing the fault tolerance in times of base stations (BSs) failures. To discover shorter routes and to shorten the route recovery time when the association property is violated, the localised-query and quick-abort mechanisms are respectively incorporated into the protocol. To further increase cell capacity and lower transmission power requirements, a dynamic cell size adjustment scheme is introduced. The protocol is free from loops, deadlock and packet duplicates and has scalable memory requirements. Simulation results obtained reveal that shorter and better routes can be discovered during route re-constructions.
Key words: ad-hoc mobile networks, associativity-based routing, neighbour-aware mobile computing.
1. The Problem and Motivation
From the best-effort-delivery type of Wireless LANs (which are basically extensions of the existing Ethernet networks running Mobile IP network protocol) to the state-of-the-art quality- of-service based Wireless ATM networks (such as SWAN-Seamless Wireless ATM Network [16–19], the “wireless last hop” from the wireline network to the mobile host has been supported through the use of wireless base stations.
Unlike these infra-structured WLANs with base-stations (BSs) providing coverage for mobile hosts (MHs), ad-hoc mobile networks do not have any access to BSs. The problem here relates to how MHs can communicate with one other, over the wireless media, without any infra-structured network component support. The most obvious problem is to devise a scheme to compute routes which can adapt well to link changes. Conventional distributed routing schemes attempt to maintain consistent routing information by performing periodic link and topology updates. These, however, are undesirable for MHs in an ad-hoc mobile network since MHs’ migrations cause frequent link changes, which result in enormous transmissions over the wireless media to propagate and update routes. This is very inefficient in an environment where radio bandwidth and battery power are scarce resources. Hence, there is a need for a new, efficient and robust routing scheme for MHs in an ad-hoc mobile network.
Parts of this Letter appeared in IEEE International Phoenix Conference on Computers & Comunications (IPCCC’96) as “A Novel Distributed Routing Protocol to Support Ad-Hoc Mobile Computing”. This invention, also named as “Cambridge Ad-Hoc Routing Protocol”, has been filed for a patent.
C.-K. Toh is supported by King’s College Cambridge and Cambridge Commonwealth Trust Scholarships. He can be reached at: Chai-Keong.Toh@cl.cam.ac.uk.
The organisation of this Letter is as follows. Section 2 discusses briefly on the existing ad-hoc mobile routing schemes while Section 3 describes the characteristics of ad-hoc mobile networks. Section 4 introduces a new concept of associativity-based routing (ABR), followed by a detailed description of the protocol in Section 5. The performance of the routing protocol is evaluated with a migration-based mobile network simulator, which is described in Section 6.
Comparison of our proposed protocol with existing routing schemes is presented in Section 7.
Finally, Section 8 reveals a dynamic cell size adjustment scheme (DCSAS) for ABR while Section 9 provides a brief discussion on ad-hoc Wireless ATM LANs and mobile applications.
2. Ad-Hoc Mobile Network Routing Schemes 2.1. EARLYMOBILEROUTINGSCHEMES
The ARPANET Packet Radio Network (PRN) [21, 30] is the earliest deployment of a regional- wide wireless data network. It has existed since the 1970s and has proven to be feasible and practical. In a PRN, all components (repeaters, terminals and stations) can be mobile or certain components can remain fixed while others are moving. There are basically two approaches to routing and packet forwarding in PRNs and since these approaches are used here as the basis of reference for ad-hoc mobile routing, they are briefly described below.
Routing in PRNs
In “point-to-point” routing, the station computes all the routing information and the decision is either distributed to the repeaters involved in the route or to the source packet radio (PR). This scheme was found to be suitable for slow moving user terminals.
However, in “broadcast routing”, each packet radiates away from the source PR with a wave-front-like propagation. Since no station needs to be present to compute routes, the destination address serves to identify the intended recipient. For fast moving user terminals, broadcast routing was found to be useful as it avoided the need to process rapidly changing routes.
Packet forwarding in PRNs
The connectionless approach to packet forwarding requires some background operation to maintain up-to-date network topology and link information in each node. This means that as topology changes, the background routing traffic can be substantial. This is commonly associated with broadcast routing, where each packet carries sufficient routing information for it to arrive at the destination. However, in the connection-oriented approach, an explicit route establishment phase is required before data traffic can be transported. This approach is commonly associated with point-to-point routing, where each node in a route has a lookup table for forwarding incoming packets to the respective out-going links. Hence, if the topology changes, a route re-establishment phase is needed. A detailed contrast between these approaches is found in [2] and [26].
2.2. CURRENTMOBILEROUTINGSCHEMES
Many ad-hoc mobile routing schemes have evolved. They include those of [3, 8, 9, 13, 20, 22, 25, 28]. Most of these schemes are based on either broadcast or point-to-point routing using either the connectionless or connection-oriented packet forwarding approach. We shall briefly describe five such schemes.
The “Layer Net” self-organising protocol proposed in [22] uses a connectionless packet forwarding approach. Broadcast routing is used for the initial network connectivity construc- tion and the subsequent topology maintenance as a result of nodes’ movements and link changes. Because routes are not constructed based on demand and topology updates have to be performed in sympathy with link changes, the overall signalling traffic can be quite substantial.
Cluster-based routing [20], in essence, uses the broadcast routing and connectionless packet forwarding approach. It relies on existing routing schemes such as link-state or distance-vector routing to derive network topology and link information. On top of this, a clustering method- ology is used to reduce the number of updates due to MHs’ migrations. Routes are constructed between all pairs of nodes and route maintenance is essentially cluster maintenance. Hence this method is inefficient.
In source-initiated distributed routing [8], however, a combination of point-to-point and broadcast routing using the connection-oriented packet forwarding approach is used. Here routes are initiated by the source and are constructed based on demand. Hence, this scheme forgoes the need to constantly propagate up-to-date routing information throughout the net- work. However, because alternate route information is used during route re-construction (RRC), problems of stale routes exist.
As for the dynamic source routing scheme proposed in [13], periodic route advertisements are avoided and route caches are used to store source routes that a mobile host has learnt over time. Routes are source-initiated and discovered via a route discovery protocol. With source routing, the sender explicitly lists the route in each packet’s header, so that the next-hop nodes are identified as the packet transits towards the destination. Because route cache information is used, accurate updates of these route caches are essential. Because the sender has to be notified each time a route is truncated, the route maintenance phase does not support fast route re-construction.
Finally, for destination sequence distance-vector (DSDV) routing [4], an enhancement to the existing distance-vector Bellman–Ford routing is made in order to support ad-hoc MHs.
Because each MH has to periodically advertise its view of the network topology, this scheme is inefficient. Similar to cluster-based routing, this scheme uses the broadcast routing and connectionless packet forwarding approach. Hence, the problems associated with existing routing schemes motivate us to seek a better routing approach to support ad-hoc mobile computing.
3. Characteristics of Ad-Hoc Mobile Networks
3.1. AD-HO CCOMMUNICATIONSERVICEDISCOVERY
Conventional computer supported groupware technology forces users to be in the same place where the collaboration facilities are installed. Users are also compelled to reserve the collabo- ration facilities well in advance to prevent usage conflicts. In addition, partners in collaboration are restricted to those who can connect to the same network. Ad-hoc mobile networks, on the other hand, allow spontaneous LANs to be created anytime and anywhere. Such networks support nomadic collaborative computing [1], which occur anytime and anywhere, without the support of BSs. Hence, users are no longer bounded by the time and place to perform collaborative computing.
For ad-hoc mobile communications to be possible, a host discovery mechanism must exist.
When an association is formed (through recognising a MH’s beacon), the mobile application is informed of the new mobile user who can then participate in nomadic collaborative computing.
This can appear as an icon which when clicked, reveals details of the new participant. The new associativity of one MH with another can also be propagated to the nodes’ neighbours, so that all other neighbouring MHs within the mobile subnet can be aware of the existence of the new user. Alternatively, an MH can choose only to be aware of its immediate neighbours and can later perform an on-demand neighbour discovery broadcast when it desires to communicate with MHs outside its vicinity. In this manner, MHs can also discover what services are available from which MHs.
3.2. TYPES OFCOMMUNICATION
An ad-hoc mobile network has no BSs to provide connectivity to backbone hosts or to other MHs. There is no need for handover and location management at all. In such a network, each MH acts as a router, forwarding packets from one MH to another. The communication connectivity is fairly “weak”, as any migration by MHs participating in one or more routes will cause the route to become invalid. The MH should not spend most of its time updating and computing routes in sympathy with MHs’ movements. Such schemes are highly impractical, inefficient and result in low data throughput. Hence, a novel routing scheme is needed to provide efficient and high throughput communication among ad-hoc MHs. Prior to that, we shall examine the possible types of ad-hoc mobile communications.
MHs in an ad-hoc mobile network can communicate with their immediate peers, i.e.
peer-to-peer, which is a single radio hop away. However, if three or more MHs are within range of each other (but not necessarily a single hop away from each other), then “remote-to- remote” MHs communications exist. Remote-to-remote communications are associated with group migrations. The traffic characteristics for different types of ad-hoc communications are discussed below.
3.3. TRAFFICPATTERNS
What are the traffic and migration patterns associated with MHs in an ad-hoc mobile network?
Depending on the MHs’ associativity and migration profiles and based on the inference that ad-hoc MHs wish to communicate when they are within range of each another, we identify the following three types of traffic patterns.
Figure 1(a) shows the network traffic for the case of peer-to-peer communications, where communications among MHs are usually between pairs that are a single radio hop away from each other. These MHs can be migrating in pairs or in groups.
The second scenario is remote-to-remote communications, where mobile users migrate in groups and communicate with destination hosts that are beyond a single radio hop away. An example of this is a group of mobile users gathering together in some foreign place for a discussion, a seminar or a mini-conference where there is no luxury of radio coverage by BSs.
Here, remote-to-remote communications allow multi-party collaboration and spontaneous ad- hoc network formation to be possible. The resulting network traffic is shown in Figure 1(b).
The third scenario refers to MHs migrating without any correlation with one another and communications can exist between or among the MHs. This is the extreme scenario where the MHs are highly dynamic, resulting in very poor connectivity and low throughput as compared
Figure 1. Traffic patterns in an ad-hoc mobile network.
to MHs migrating in groups. Hence, the network traffic may consist of a series of abrupt bursts, as shown in Figure 1(c).
3.4. INTEGRATION WITHBS-ORIENTEDWLANS
It is desirable for a MH to be able to function in both ad-hoc or BS-oriented WLAN environments.1When a MH receives beacons generated by other MHs, it automatically invokes ad-hoc routing to support mobile-to-mobile communication. However, when it receives bea- cons generated by BSs, the MH knows that it has access to the wired network and hence conventional routing protocols supported by location management, registration, handovers, etc., can be invoked.
Mobile applications can also be made intelligent enough to decide which communication mode, i.e. ad-hoc or BS-oriented, best suits the service requirements. It shall be shown later that both ad-hoc and BS-oriented modes can be combined to provide fault tolerance against BSs’ failures.
3.5. TYPES OFMH MOVEMENTS
3.5.1. Movements by SRC, DEST and INs
A route in an ad-hoc mobile network comprises of the source (SRC), destination (DEST) or/and a number of intermediate nodes (INs). Movements by any of these nodes will affect the validity of the route.
A SRC in a route has a downstream link and when it moves out of its downstream neigh- bour’s radio coverage range, the existing route will immediately become invalid. Hence, all the downstream nodes may have to be informed so as to erase their invalid route entries.
Likewise when the DEST moves out of the radio coverage range of its upstream neighbour, the route becomes invalid. However, unlike the SRC, the upstream nodes will have to be informed so as to erase their invalid route entries.
1 This, in fact, is one of the functional specifications laid down by the IEEE 802.11 committee.
Figure 2. Ad-hoc mobile subnets merging and fragmenting.
Lastly, similar to the SRC and DEST nodes’ movements, any movements by one of the INs supporting an existing route may cause the route to become invalid.
All these nodes’ movements cause conventional distributed routing protocols to respond in sympathy with the link changes, in order to update all the remaining nodes within the network, so that consistent routing information can be maintained. However, this involves broadcasting over the wireless medium which results in wasteful bandwidth and an increase in the overall network control traffic. Consequently, a better route re-construction scheme is required.
3.5.2. Movements by Subnet-Bridging MH
In addition to the above-mentioned moves, moves by a MH which is performing subnet- bridging function between two virtual mobile subnets can fragment the virtual subnet into smaller subnets. Likewise, some MHs can cause subnets to be merged, resulting in bigger subnets. This is illustrated in Figure 2. When the mobile subnets merge to form bigger subnets, the routing algorithm can accept the new subnet by updating all the nodes’ routing tables (RTs).
This is however very inefficient. In our routing protocol, we forgo this inefficient scheme and only choose to update the affected MHs’ associativity tables, which is already an inherent part of the MH’s radio data-link layer functions. Likewise, this applies to partitioning subnets.
The property of a mobile subnet states that if both the SRC and DEST are elements of the subnet, a route or routes should exist unless the subnet is partitioned by some subnet- bridging MHs. Mobile subnets can be used to support nomadic collaborative computing, and the collaboration partners can grow in size when two collaboration groups join or when new mobile users join by coming into range.
3.6. CONCURRENTMHs’ MOVEMENTS
In reality, concurrent moves by SRC, INs and DEST exist. Hence, consistency is needed to ensure that multiple route re-constructions or updates will ultimately converge and no deadlock or stale routes will exist.
Figure 3. Time and spatial representation of associativity of a MH with its neighbours.
4. Associativity-Based Routing (ABR)
ABR is a compromise between broadcast and point-to-point routing. Similar to [8], ABR only maintain routes for sources that actually desire routes. However, ABR does not employ route re-construction based on alternate route information stored in INs (thereby avoiding stale routes). In addition, routing decisions are performed at the DEST and only the best route will be selected and used while all other possible routes remain passive. This, therefore, avoids packet duplicates. Furthermore, the selected route tends to be more long-lived due to the property of associativity, which is described below.
4.1. RULE ANDPROPERTY OFASSOCIATIVITY
4.1.1. Rule of Associativity
This rule states that a MH’s association with its neighbour changes as it is migrating and its transiting period can be identified by the associativity “ticks”. The migration is such that after this unstable period, there exists a period of stability, where the MH will spend some dormant time within a wireless cell before it starts to move again.2The threshold where the associativity transitions take place is defined byAthreshold, as shown in Figure 3.
2 Refers to moves causing link changes with the MH’s neighbours.
Figure 4. Dormant time distribution of 52 badge wearers in a day at the Computer Laboratory.
Table 1. Dormant time distribution of 52 badge wearers in a week at the computer laboratory.
Dormant time (minutes)
Distributions Day of the week
Mon Tue Wed Thur Fri
Maximum 299.15 277.00 281.68 223.06 297.64
Minimum 5.08 5.06 5.10 5.01 5.02
Mean 35.79 36.26 41.08 40.84 47.99
Standard deviation 46.63 50.88 50.55 55.40 62.81
Associativity ticks are updated by the MH’s data-link layer protocol, which periodically transmits beacons identifying itself and constantly updates its associativity ticks in accordance with the MHs sighted in its neighbourhood. In a scenario where an ad-hoc WLAN has a wireless cell size of 10 m with a MH’s minimum migration speed of 2 m/s and a beacon transmission interval of a second, the maximum possible associativity ticks of the migrating MH with its neighbours is five. Likewise, the neighbouring MHs will also record associativity ticks of no more than five. This value isAthresholdand any associativity ticks greater than this threshold implies periods of association stability.
To further support our claim that a mobile user will have some dormant time at a location before it starts to move again, we gathered the mobility traces of 52 badge wearers from the Active Badge System3 for five consecutive days (from 8 am till 6.30 pm) at the Cambridge University Computer Laboratory.4 The traces provide specific information on the sighted location (domain and network identifiers), the wireless cell (badge sensor identifier), the MH (the badge identifier) and the time when the MH is sighted (time-stamps).
Figure 4 and Table 1 show the dormant time distributions obtained for a day and a week respectively. The average dormant time ranges from 35.79 to 47.99 mins. Hence, we believe that a practical mobile user will spend some dormant time at a location before he/she decides to move again.
3 Details available from http: //www.cl.cam.ac.uk/abadge/documentation/abwayin.html.
4 Details available from http: //www.cl.cam.ac.uk/site-maps/site-maps.html.
Figure 5. Interlocking phenomenon in ad-hoc mobile networks.
4.1.2. Property of Associativity
A MH is said to exhibit a high state of mobility when it has low associativity ticks with its neighbours. On the other hand, if high associativity ticks are observed, the MH is in the stable state and this is the ideal point to select the MH to perform ad-hoc routing. Consequently, if all the MHs in a route have high associativity ticks, an inter-locking phenomenon arises where
“my” degree of associativity ticks will be high if “you” do not move out of reachability (i.e.
symmetric mutual-dependent property) and are in stable state, as illustrated in Figure 5. The associativity ticks are reset when the neighbours or the MH itself moves out of proximity, not when the communication session is completed and the route made invalid.
4.1.3. Applicability to BS-Oriented WLANs
The properties of associativity can also be applied to BS-oriented WLANs. When a MH
“sees” a BS, its associativity ticks with the BS will be high. But these associativity ticks will be reset on BS’s failure (equivalent to an associated node moving away). Hence, under such circumstances, the MH can apply associativity-based ad-hoc routing to re-route its packets to its neighbouring MHs who may have access to other BSs. In this manner, robustness can be achieved during BSs’ failures.
4.2. NEWROUTINGMETRICS
Conventional routing qualities are characterised by:
Fast adaptability to link changes (recovery time).
Minimum hop path to destination.
Propagation delay.
Loop avoidance.
Link capacity.
However, is fast adaptability to changes in network topology a plus? As mentioned earlier, some protocols go to the extreme of frequent broadcasts to attain fast route convergence.
In an ad-hoc mobile network, fast adaptability at the expense of excessive radio bandwidth consumption is undesirable. The qualities of a good route should not only include the number of hops and the round trip propagation delay. We identify the following new routing metrics:
Table 2. The ABR route selection algorithm.
Let
S
ibe the set of possible routes from SCR!DEST, wherei
=1;
2;
3::::
Let RLijbe the relaying load in each node
j
of a route inS
i, wherej
=1;
2;
3::::
Let RLmaxbe the maximum route relaying load allowed per MH.
Let ATthresholdbe the min. associativity ticks required for association stability.
Let ATijrepresent the associativity ticks in each node
j
of a route inS
iLet
H
irepresent the aggregate degree of association stability of a route inS
i.Let
L
irepresent the aggregate degree of association instability of a route inS
i.Let
H
iaverepresent the average degree of association stability of a route inS
i.Let
L
iaverepresent the average degree of association instability of a route inS
i.Let
Y
irepresent the number of nodes of a route inS
ihaving acceptable route relaying load.Let
U
irepresent the number of nodes of a route inS
ihaving unacceptable route relaying load.Let
Y
iaverepresent the average acceptable route relaying load factor.Let
U
iaverepresent the average unacceptable route relaying load factor.Begin
For each route
i
inS
iBegin
a
0For each node
j
in routeS
iBegin
If(ATijATthreshold)
H
i++;else
L
i ++;If(RLijRLmax)
U
i++;else
Y
i ++;a
++;End
H
iave=H
i=a
;L
iave=L
i=a
;U
iave=U
i=a
;Y
iave =Y
i=a
;End
Best Route Computation
Let the set of acceptable routes with
U
iave=0 andH
iave 6=0 beP
l, whereP
lS
iBegin
Find Route With Highest Degree of Association Stability Compute a route
k
fromP
lwithH
kave> H
lave, 8l
6=k
;or if a set of routes
K
nexists such thatH
K1ave=H
K2ave=H
Kpavewhere
n
=f1;
2;
3;:::;p
gBegin
Compute Minimum Hop Route Without Violating Relaying Load Compute a route
K
kfromK
nwith MinfK
kg<
MinfK
mg, 8m
6=k
;or if a set of routes
K
oexists such that MinfK
1g=MinfK
2g=MinfK
qg,where
o
=f1;
2;
3;:::;q
gBegin
Multiple Same Associativity & Minimum Hop Routes Exists Arbitrarily select a minimum hop route
K
kfromK
oEnd End End End
Longevity of a route.
Relaying load of INs supporting existing routes.
Knowledge of link capacities of the selected route.
The “longevity” of a route is important as the merits of a shorter but short-lived route will be denigrated due to frequent data flow interruptions and the need for route re-constructions.
In addition, even relaying load distribution is important in an ad-hoc mobile network, as no one particular MH should be unfairly burdened to support many routes and perform many packet-relaying functions. This also alleviates the possibility of network congestion. Finally, since the associativity of a MH with its neighbours also reflects the number of contenders within a wireless cell, the approximate aggregated throughput for the selected route can be made known to the mobile user prior to transmission. This, therefore, allows the user to either proceed with or abort the transmission.
4.3. ROUTESELECTIONRULES
Given a set of possible routes from the source node (SRC) to the destination node (DEST), if a route consists of MHs having high associativity ticks (therefore indicating spatial and connection stability), then that route will be chosen by the DEST, despite other shorter hop routes. However, if the overall degree of association stability of two or more routes are the same, then the route with the minimum hops will be chosen. If multiple routes have the same minimum-hop count, then one of the route is arbitrarily selected. The ABR route selection algorithm, which is executed at the DEST, is formally stated in Table 2.
5. ABR Protocol Description
The associativity-based routing protocol consists of three phases, namely
Route discovery phase.
Route re-construction (RRC) phase.
Route deletion phase.
Initially when a source node desires a route, the route discovery phase is invoked. When the link of an established route changes due to SRC, DEST, INs or subnet-bridging MHs’ migration, the RRC phase is invoked. When SRC no longer desires the route, the route deletion phase is initiated. These three phases will be discussed below. A summary of the characteristics of the routing protocol is presented in Table 11.
5.1. ROUTEDISCOVERYPHASE
The route discovery phase allows an approximation of the data throughput associated with the selected route to be computed. This is achieved through the knowledge of associativity ticks of neighbours in the route and the relaying load of nodes supporting the route. The route discovery phase consists of a broadcast query (BQ) and an await reply (REPLY) cycle, which is described below.
BQ-REPLY cycle
Initially all nodes except those of DEST’s neighbours, have no routes to the DEST. A node desiring a route to the DEST broadcasts a BQ message, which is propagated throughout the ad-hoc mobile network in search of MHs which have a route to the DEST. Here, a sequence
Table 3. BQ control packet.
TYPE SRC ID DEST ID LIVE IN IDs METRICS SEQ NO. CRC
Figure 6. Updating associativity metric during BQ packet propagation.
number is used to uniquely identify each BQ packet and no BQ packet will be broadcast more than once.
Once the BQ query has been broadcast by the SRC, all INs that receive the query will check if it has previously processed the packet. If affirmative, the query packet will be discarded, otherwise the node will check if it is the DEST. If it is not the DEST, the IN appends its MH address/identifier at the IN IDs field of the query packet and broadcasts it to its neighbours (if it has any). The associativity ticks with its neighbours will also be appended, along with its relaying load, link propagation delay and the hop count.
The next succeeding IN will then erase its upstream node’s neighbours’ associativity ticks entries and retain only those concerned with itself and its upstream node. In addition, because of the association ticks symmetry between nodes, the associativity ticks received from the upstream node can be checked for validity. In this manner, the query packet reaching the DEST will only contain the intermediate MHs’ addresses (hence recording the path taken) and their associativity ticks (hence recording the stability state of the INs supporting the route) and the route relaying load, together with information on route propagation delays and hop count. The process is illustrated in Figure 6. The resulting BQ packet is variable in length and its format is shown in Table 3.
The DEST will, at an appropriate time after receiving the first BQ packet, know all the possible routes and their qualities. It can then select the best route (based on the selection criteria mentioned earlier) and send a REPLY packet back to the SRC, via the route selected.
This causes the INs in the route to mark their routes to DEST as valid and this means that all other possible routes will be inactive and will not relay packets destined for the DEST, even if they hear the transmission. This, therefore, avoids duplicated packets from arriving at the DEST.
Table 4. REPLY control packet.
TYPE SRC ID DEST ID INs IDs SEQ ID ROUTE QUALITIES CRC
Figure 7. REPLY interruption caused by unexpected INs’ movements.
While the BQ query packet propagates to the DEST, each node relaying the BQ packet will know its hop count from the SRC. Likewise, when the REPLY packet propagates back to the SRC, the INs can also compute their distances to the DEST. The REPLY packet is variable in length and has the format shown in Table 4.
Case when SRC never receives REPLY
There may be some rare instances when the SRC never receives DEST’s REPLY because of some unexpected “not-yet-selected” INs’ movement. In such circumstances, the SRC will eventually BQ TIMEOUT and sends another BQ query. Since the downstream neighbour of the migrating IN realises the associativity change, it will send a RN[STEP=1]5 (Route Notification) packet in the downstream direction, deleting all the downstream nodes’ invalid routing table entries. Another situation occurs when a selected IN moves while the REPLY propagation is still in progress. The upstream neighbour of the migrating node will perform a LQ[H]6 (Localised Query) process to discover a new partial route, while the downstream neighbour sends a RN[1] packet towards the DEST, thereby erasing all invalid downstream nodes’ routing entries. Hence, while the RRC is in progress, the REPLY packet continues to propagate towards the SRC. Figure 7 illustrates these two scenarios.
5.2. ROUTERE-CONSTRUCTION(RRC) PHASE
The route maintenance phase performs the following operations :
Partial route discovery.
Invalid route erasure.
Valid route update.
New route discovery (worst case).
These operations may be invoked by any of the four moves mentioned earlier. Before concurrent moves are analysed, it is essential to examine the consequence of individual node’s movements. In the proposed routing protocol, the selected route is more likely to be long-lived due to the property of associativity. However, if unexpected moves do occur, the protocol will
5 RN[1] will be explained in the next Section.
6 LQ[H] will be explained in the next Section. Note that “H” refers to “LIVE” of BQ and LQ packets.
Figure 8. Route maintenance when SRC, DEST and IN moves.
Table 5. RN control packet.
TYPE ORG ID SRC ID DEST ID SEQ ID STEP DIR CRC
attempt to quickly locate an alternate valid route without resort to broadcast query, unless necessary. The following narrations shall refer to Figures 8a, b and c respectively.
5.2.1. SRC Node Movements
Since the routing protocol is source-initiated, any moves by the SRC will invoke an RRC process equivalent to that of a route initialisation, i.e. via a BQ REPLY process. It will be shown later that this avoids “multiple-RRCs” conflicts as a result of concurrent nodes’
movements.
5.2.2. DEST node movements
When the DEST moves, the DEST’s immediate upstream neighbour (i.e. the pivoting node) will erase its route. It then performs a LQ[H] process to ascertain if the DEST is still reachable.
“H” here refers to the hop count from the upstream node to the DEST. If the DEST receives the LQs, it will select the best partial route and send a REPLY, otherwise the LQ TIMEOUT period will be reached and the pivoting node will backtrack to the next upstream node.
During the backtrack, the new pivoting node will erase the route through that link and perform a LQ[H] process until the new pivoting node is greater than half hopsrc dest away from the DEST or when a new partial route is found. If no partial route is found, the pivoting node will send a RN[1] packet back to the SRC to initiate a BQ process. While the RN packet is fixed in length, the LQ packet is not. The formats of the RN and LQ packets are shown in Tables 5 and 6.
The ORG ID is the pivoting node ID while the SRC and DEST IDs identify the route.
STEP=0 in the RN control packet means that the backtracking process is to be performed one hop at a time (in the upstream direction) while a STEP=1 implies that the RN packet will be
Table 6. LQ control packet.
TYPE SCR ID DEST ID INs IDs METRICS SEQ ID LIVE CRC
propagated straight back to the SRC to invoke a BQ process or to the DEST to erase invalid routes. The DIR flag serves to indicate the direction of RN[1] propagation.
5.2.3. Intermediate Node (IN) Movements
Upper arm INs’ moves
The “upper arm” of a route refers to the INs and the DEST that contribute to half the route length from SRC to DEST. When any IN moves, its immediate upstream node (i.e. the pivoting node) removes its outgoing node entry and its immediate downstream neighbour propagates a RN[1] packet towards the DEST, thereby deleting all the subsequent downstream nodes’
invalid routing entries. A new partial route to the DEST needs to be found.
A LQ[H] process is then invoked by the pivoting node to locate alternate partial routes.
The DEST may receive multiple LQs, hence it needs to select the best partial route and return a REPLY to the pivoting node. This causes all INs between DEST and the pivoting node to update their RTs. On receiving the REPLY, the pivoting node updates its RT entries and appends the next hop (outgoing) node ID into the data packet. This ensures that only one partial route is selected.
As before, the LQ[H] process is performed based on a suitable H value. If the pivoting node is X hops away from the DEST via the previous active route, then H=X will be used in the hope that the DEST is still within X hops range (reachable via other paths) or shorter. This, therefore, attempts to rebuild partial paths of equal or shorter lengths (i.e. route optimisation during RRCs).
However, if no partial route exists, LQ TIMEOUT will expire and a RN[0] packet will be sent by the pivoting node to the next upstream node, and the cycle repeats until the next pivoting node has a hop count greater than half hopsrc destor when a new partial route to the DEST is found.
Lower arm INs’ moves
The “lower arm” refers to the SRC and INs that contribute to half the route length from SRC to DEST. If any of these nodes moves, RN[1] packet will be propagated downstream towards the DEST, and the pivoting node will perform LQ[H] and await the DEST’s REPLY. If no REPLY is received, a RN[0] packet is sent to the next upstream node and the new pivoting node then invokes the LQ[H] process again, but with a different value of H. The cycle proceeds until the new pivoting node is the SRC (where the BQ process will be initiated to discover a new route) or a partial route to the DEST is found.
5.2.4. Subnet-Bridging MH Movements
The migration of a subnet-bridging MH beyond the radio coverage of its neighbouring MHs will cause the mobile subnet to be partitioned. If an existing route does not span across the fragmented subnets, the route is not affected and only the subnet-bridging MH’s upstream and downstream neighbours need to update their route and associativity entries. All other MHs remain ignorant and do not perform any route updates.
However, if existing routes span across subnets (i.e. the subnet-bridging MH is an IN of the route), then the route is invalidated as the DEST is no longer reachable, despite any LQ or BQ attempts. Under such circumstances, the LQ RN cycle will eventually inform the SRC about the partitioning and the SRC can then invoke BQ query several times or it can inform the mobile user about the partitioning and prompt him to try later.
5.2.5. Concurrent Nodes Movements
Race conditions exist due to multiple invocations of RRC processes as a result of concurrent movements by SRC, DEST and INs. The following explains why the proposed routing protocol is immune to “multiple-RRCs” conflicts and how one RRC is valid ultimately.
DEST-moves RRC interrupted by upstream INs’ moves
When the DEST moves and while the RRC is in progress, any upstream INs moves will cause their respective downstream neighbours’ route to be deleted. The new pivoting node nearest to the SRC will perform the RRC and all other RRCs will be passive when they hear the newer LQ broadcast for the same route. Hence, only one RRC is valid.
Upper-arm IN RRC interrupted by lower arm INs’ moves
This is the same as the above-mentioned case. Note that the same argument can be applied to the case when a LQ process has to be aborted and a RN[1] packet has to be sent to the SRC to invoke a BQ but is hindered due to some upstream INs’ movements. The new pivoting node nearest to the SRC will swamp the earlier RRC processes by invoking a new LQ.
Lower-arm IN RRC interrupted by upper arm INs’ moves
While a lower arm IN RRC is taking place, any movements by any upper arm INs will not result in a LQ[H] or RN[1] process being initiated since the lower arm IN has earlier sent RN[1] packet downstream to erase invalid routes. If the RN[1] packet does not succeed in propagating towards the DEST, the LQ[H] process initiated by the lower arm IN will also serve to delete these invalid routes.
Lower/upper-arm IN RRC interrupted by DEST’s moves
This has no effect on the RRC, as the LQ[H] process uses a localised query approach to locate the DEST. Once the DEST is in its stable state and is reachable from the pivoting node, the RRC process will be successful.
Lower/upper-arm IN RRC interrupted by SRC’s moves
While lower or upper arm IN RRC is in progress, any moves by the SRC will result in a BQ, which will swamp out all on-going LQ REPLY RN processes related to that route.
Hence, unfruitful and stale RRCs will not continue and a new route has to be discovered via the BQ process.
SRC and DEST nodes moving away from INs
When this occurs, RRCs as a result of DEST and SRC moves will be initiated. However, the BQ process initiated by the SRC will again swamp out all unnecessary on-going RRCs.
DEST migrating into SRC’s radio coverage range
When the DEST migrates, RRC is achieved via the LQ[H] process. However, when the DEST is within the SRC’s radio coverage range, packet duplicates will result at the DEST since the DEST now receives packets from the SRC directly and also from the original SRC-DEST route. Hence, to avoid packet duplicates and non-optimal routes, the SRC, on discovering that the DEST is within range and is in stable state, will send a RN[1] packet downstream to erase existing route and will re-establish a new single hop route with the DEST.
Table 7. RD Control packet.
TYPE SRC ID DEST ID LIVE SEQ NO. CRC.
5.2.6. LQ-REPLY Cycle Interruption
During a LQ-propagation and REPLY-await process, if any of the upstream nodes7(i.e. lower arm INs) break up, a RN[1] packet will be propagated downstream, erasing all the downstream INs’ routes entries. The existing pivoting node will ignore any subsequent REPLY to its LQ.
The new pivoting node will resume with a new LQ-REPLY process.
5.2.7. Route Erasure and Updates
In ABR, no attempt is made to retain alternate routes, as maintaining them causes overhead.
Only one route will be selected and only one route is valid for a particular route request. The avoidance of using alternate routing information means that problems associated with looping due to INs having stale routes are absent and there is no need for periodic network-wide broadcast and route updates. This is, however, not the case for the source-initiated distributed routing scheme proposed in [8].
Any alternate route will have to be discovered via a LQ or BQ process, which may give rise to better (shorter hop and long-lived) routes. The DEST, on receiving multiple LQs, will select the best route and reply to the SRC. During the LQ REPLY RN cycle, invalid INs routes are erased by RN[1] packets and INs forming the new partial route will have their route entries updated when they relayed the REPLY packet from the DEST to the pivoting node. If the LQ REPLY RN cycle fails, the subsequent new pivoting node will have its route entries erased by RN[0] packet during the backtrack process. If all the possible backtrack LQ REPLY RN cycles fail, all the upstream nodes will have their route entries erased via RN[0] and RN[1]
packets and the SRC will then revert back to the BQ REPLY cycle.
Finally, for the case of BQ query, any INs receiving a BQ and having invalid routes will result in these routes being erased, therefore ensuring that no invalid routes exist in the INs.
5.3. ROUTEDELETIONPHASE
When a discovered route is no longer desired, a route delete (RD) broadcast will be initiated by the SRC so that all INs will update their routing table entries. A full broadcast is used compared to directed broadcast. This is so because the nodes in a route change during route re-constructions, hence using directed broadcast will be unsuitable unless the SRC is always informed about any changes to the route path. Similar to BQ, the RD control packet has a LIVE=1to achieve a full wave-like broadcast. Its format is shown in Table 7.
5.4. PACKETHEADER
Since a long packet header results in low channel utilisation efficiency, in our proposed protocol, each data packet header will only contain the neighbouring node routing information, not all the nodes in the route. Each IN will renew the next-hop information contained in the header before propagating the packet upstream or downstream. Hence, a hybrid routing scheme
7 Downstream nodes’ movements are not a concern in this case.
Table 8. Packet header.
Routing header field Function
SRC ID Packet forwarding
DEST ID Route identification
Sequence no. Duplicates prevention, uniqueness Service type Packet priority
Last IN Passive acknowledgement Next IN Duplicates prevention, routing Current IN Acknowledgement, routing
Table 9. Routing table.
Destination Source Incoming IN Outgoing IN Distance
N
aN
xN
zN
j 4N
kN
yN
iN
o 3Total no. of active routes supported (relay load): 2
which is a combination of broadcast and point-to-point routing is used. The purpose of the individual fields of the packet header is summarised in Table 8.
5.5. DATAFLOWACKNOWLEDGEMENT
As in PRNs, we employ a passive acknowledgement scheme for packets in transition. When a node receives a packet and performs relaying via a radio transmission to its neighbours, its previous neighbour that has sent it the packet will have heard the transmission and hence this is indirectly used as an acknowledgement to the packet sent. On the other hand, active acknowledgements will only be sent by the DEST as it no longer has a neighbour to relay the packet to. Hence, this provides a data flow acknowledgement mechanism for packet forwarding in an ad-hoc mobile network, which is not present in any of the existing ad-hoc mobile routing schemes.
5.6. PACKETRETRANSMISSION
While the data flow acknowledgement scheme allows forwarded packets to be acknowledged, there are situations where the acknowledgements never reach the intended receiver. This can be a result of radio interference which causes a sudden loss of radio connectivity. Hence, if a MH has forwarded a packet and does not receive an acknowledgement within a certain time interval, it retransmits the packet forXtimes, after which the neighbouring MH will be considered as out-of-reach and RRC procedures will be invoked. If however, the radio link returns before the retransmission counter expires, the packet forwarding process continues, as illustrated in Figure 9.
5.7. ROUTINGTABLE(RT)
The RT of a node supporting existing routes is shown in Table 9. The table reveals that every node supporting on-going routes will map incoming packets from a particular upstream
Table 10. Neighbouring table.
Neighbours Associativity ticks Link delays (units) (msecs)
N
a 5 75N
b 15 50Figure 9. Handling packet retransmission in ABR.
node to the corresponding out-going downstream node. Every node will also keep track of its distance (hop count) to the DEST and a record of the total routes that it is currently supporting.
5.8. NEIGHBOURINGTABLE(NT)
The NT is usually updated by the data-link layer protocol, which will generate, receive and interpret beacons from the neighbouring MHs or BSs and pass this information up to the higher protocol layers. Nomadic collaborative applications can then utilise the NT information to update their participants’ present and absent lists. The structure of a NT is shown in Table 10.
5.9. CONTROLPACKETS“SEEN” TABLES
While the BQ query process is activated via a radio broadcast, the LQ query process is invoked via a localised broadcast. To avoid MHs from processing and relaying the same BQ or LQ packet twice, BQ and LQ “seen” tables are needed. If the received control packet type, route identifier and sequence number match an entry in the “seen” table list, then the packet is discarded. On the other hand, because the REPLY and RN control packets utilise “directed”
broadcast (since intended recipients’ addresses are contained in the control packet), “seen”
tables for these packets are not necessary.
5.10. ABR PROTOCOLSUMMARY
Table 11 summarises the procedures of the routing protocol under different MHs’ associativity states. The outstanding feature is that no RRCs is needed so long as the property of associativity
Table 11. Summary of route reconstructions under varying MHs’ migrations.
Associativity Associativity violated
valid INs & DEST moves SRC moves Subnet bridging MH moves Concurrent moves
Normal Worst Route within Route spans
case case subnet across subnets
No route LQ, BQ, BQ, No route Network is Ultimately
reconstructions REPLY REPLY REPLY reconstructions partitioned only one route
are needed cycle cycle cycle are needed BQ REPLY reconstruction
success success success cycle will cycle is valid
retry before aborting
Table 12. Complexity comparison of routing protocols.
Routing protocol TC CC
ILS
O
(d
)O
(2E
)MS
O
(d
2)O
(N
2)DBF (link failures)
O
(N
)O
(N
2)NP (QRY-RPY, initialisation)
O
(2d
)O
(2N
)NP (QRY-RPY, postfailure)
O
(2l
)O
(2x
)NP (FQ-RPY, postfailure)
O
(2l
)O
(2x
)ABR (BQ-RPY, initialisation)
O
(d
+z
)O
(N
+y
)ABR (LQ-RPY, postfailure)
O
(l
+z
)O
(x
+y
)Flooding (no control packets) 0 0
interlock remains valid. When this property is violated, the protocol will invoke a LQ or BQ process to quickly locate alternate routes. To further illustrate the simplicity and efficiency of the ABR protocol, we compare our ABR protocol with several existing routing algorithms in terms of time and communication complexities.8Table 12 (extracted with permission from [8]) reveals that ABR is simpler and more efficient than all other algorithms compared. The abbreviations used in Table 12 are listed below:
N = number of nodes in the network,
E = number of links in the network,
d = network diameter,
x = number of nodes affected by a topological change,
l = diameter of the affected network segment,
y = total number of nodes forming the directed path where the REPLY packet transits,
z = diameter of the directed path where the REPLY packet transits, ILS = idealised link–state protocol,
MS = Merlin–Segall routing algorithm [23], DBF = distributed Bellman–Ford algorithm,
NP = routing protocol proposed in [8], ABR = associativity-based routing.
8 Time complexity refers to the number of steps required to perform a protocol operation. Communication complexity refers to the number of messages exchanged in performing a protocol operation.
6. Migration-Based Simulation and Performance Evaluation 6.1. SIMULATIONMODEL ANDENVIRONMENT
In order to evaluate the performance of the routing protocol during cases when the “rule of associativity” is violated (though this can be less frequent, as verified from the badge trace findings), the routing protocol is implemented within a migration-based ad-hoc mobile network simulator.
Unlike other ad-hoc mobile network models proposed in [7] and [20], our model is migra- tion based, where MHs are laptops or personal digital assistants (PDAs) users in an in-building environment. The network simulator models MHs migrating in two dimensional space with a wireless cell size of 10 m in diameter. The simulator randomly allocatesN MHs (where
N=30) on a 400 m2space. The resultant ad-hoc mobile network connectivity is determined from the overlapping of these wireless cells, i.e. if the shortest distance between MHs is less than the cell radius, these MHs are connected to one another. To evaluate the routing protocol for all possible source-initiated routes, i.e.N(N 1), the resultant ad-hoc network so formed must not be partitioned.
For each of the possible routes, each IN belonging to the route is progressively moved away (i.e. moved out of connectivity range with its upstream or downstream neighbours), thereby invalidating the route. RRC procedures are then invoked to locate alternate routes.
The average neighbouring factor (N.F.)9 of a route defines the association of a route with surrounding MHs and is given by:
N:F:=
N
X
i=1
nf
i
N
;
where
nf
i
=
no. of neighbours associated with nodei maximum possible no. of neighbours for a node
:
In the simulator, the maximum possible number of active neighbours associated with a node is 10. To evaluate the protocol behaviour for routes having different N.F., RRCs are performed for each of the possible N.F. till its maximum. This is achieved by gradually moving “free-nodes” (nodes that are not part of and neighbours of the route) into the coverage range of the SRC, DEST and INs. To ensure that the N.F. is1, the number of neighbours associated with a node on the route must be within the specified maximum. Hence, a node cannot be moved to the vicinity of the considered node if it causes neighbours overflow in any of the other neighbouring nodes’ cells. Figures 10 and 11 show an ad-hoc mobile network connectivity state, using the default nodes allocation seed, before and after nodes’ migrations.
The route statistics associated with the initial ad-hoc mobile network is shown in Figure 12.
6.2. SIMULATIONRESULTS ANDOBSERVATIONS
To derive an overall picture of the protocol performance, RRCs are performed for all possible source-initiated routes. To ensure that the results obtained are valid for different initial network nodes’ location and connectivity states, different possible nodes’ allocation seeds for the
9 This should not be confused with the degree of associativity (association stability) of MHs participating in the route.
Figure 10. Ad-hoc mobile network topology before nodes’ migrations.
Figure 11. Ad-hoc mobile network topology after nodes’ migrations.
random generator are used. For each of the re-constructed route, the following parameters are recorded
Route length difference (w.r.t. INs’ migrations and route N.F.).
LQ difference (w.r.t. INs’ migrations and route N.F.).
Maximum no. of LQs performed (w.r.t. INs’ migrations and route N.F.).
Figure 12. Routes statistics for ad-hoc mobile network using default nodes allocation seed.
6.2.1. Results for a Single Ad-Hoc Mobile Network
LQ difference
As shown in Figure 13, the minimum possible LQ difference is one since the DEST node is not moved but its upstream nodes. This represents 42.96 percent of the total population since 57.04 percent has LQ differences of greater than one. This implies that the actual number of LQs performed before a successful RRC is less than its worst possible maximum. Note that the results of LQ difference are derived from routes having three or more hops only. Although it is not obvious from Figure 13 that LQ difference increases with increasing N.F., our study on the raw results for routes with similar hop count and IN moves but with different N.F., confirms this point. We found that the average N.F. threshold where LQ difference began to improve occurred at 0.67.
Maximum no. of LQs performed
Figure 14 clearly reveals that the maximum number of LQs required before a successful RRC decreases with increasing N.F.. This is due to the exploitation of locality, i.e. increasing the route N.F. increases the probability of a LQ query success. To avoid being misled by the result (due to routes having hop counts of one or two), the network route types are computed (see Figure 12) and clearly the cumulative percentage of routes having 3 ! 7 hops are greater than those having one or two hops. A zero maximum LQ is due to one and two hops routes and for cases when the migrated node is equal to half hop or (route hop count 1). From the raw results, the average N.F. threshold which resulted in reductions on the maximum number of LQs performed was found to be 0.66.
Path difference
The difference between the original route length and the new route length (after re-construction) yields the path difference. In the simulator, “LIVE” is set to the hop count from the new piv- oting node to the DEST. This implies that none of the discovered route will be longer than
Figure 13. Results of LQ difference for all INs’ moves and route reconstructions with varying N.F.
Figure 14. Maximum LQ performed for all INs’ moves and route re-constructions with varying N.F.
the original route, as verified by results shown in Figure 15. Note that one and two hop routes do not contribute to the result. Out of a total of 10,438 RRCs, 52.31 percent have shorter routes while 47.69 percent have the same route length, implying that the LQ RRC process can discover shorter new routes. In addition, as the N.F. is increased, the path difference becomes significant. Again from the raw results, we computed the average N.F. threshold which caused improvements in the path difference and this value was found to be 0.74.
Figure 15. Results of path difference for all INs’ moves and route re-constructions with varying N.F.
Figure 16. Three-dimensional plot of route characteristics for 113 ad-hoc mobile networks under tests.
6.2.2. Overall Results for Different Ad-Hoc Mobile Networks
To ensure that the deductions derived from the above simulations are valid for ad-hoc mobile networks having different initial nodes’ location and connectivity states, a further 113 simu- lations (a total of 1,155,648 RRCs) are performed for all possible seeds in the 0!800 range.
The following results are recorded:
Overall routes’s characteristics
Figure 17. Two-dimensional plot of route characteristics showing mean values and 95% confidence intervals.
Figure 18. Zero path difference frequency distribution for 113 ad-hoc mobile networks.
The resulting route types for each of the possible ad-hoc mobile networks are plotted to give an overall view of the networks’ route characteristics. Both Figures 16 and 17 show that for every ad-hoc mobile network under test, the route characteristics are different. In particular, minor variations exist over routes with larger hops while considerable variations exist for routes in the 1!10 hops range. This is due to the constraint imposed by the number of MHs confined in a 400 m2physical space. The 95 percent confidence intervals and mean percentage for different route type occurrences are shown in Figure 17. Note that routes with eight hops or more have positively-skewed frequency distributions while others have normal distributions.
Figure 19. Positive path difference frequency distribution for 113 ad-hoc mobile networks.
Figure 20. Three-dimensional plot of LQ differences for 113 ad-hoc mobile networks.
Overall path differences
Figures 18 and 19 reveal that for each of the 113 different ad-hoc mobile networks (with every possible route undergoing INs’ movements and hence invoking RRC processes over increasing route N.F.), the percentage of the resultant path being shorter than that of the original path has a mean of 61.65 percent, while those having the same path length is 38.34 percent. Hence, the protocol’s nature of exploiting the advantages of locality via localised query has allowed the formation of shorter new paths.
