Computer Science Technical Report Archive

USC Computer Science Technical Reports, no. 803 (2003)

(USC DC Other)

USC Computer Science Technical Reports, no. 803 (2003)

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Ahmed Helmy – USC – July 2001 (All rights reserved) 1 NSF CAREER Proposal: Project Summary
Adaptive Architectures for Multicast Service Support in Large-Scale Mobile Ad Hoc Networks:
Design and Evaluation Framework
Ahmed Helmy , Electrical Engineering Department, University of Southern California
This proposal outlines a comprehensive career plan for advancement of research and education in the
field of wireless mobile ad hoc networks. The research plan consists of two complementary parts. The first
is to design and develop a novel architecture for scalable multicast routing in wide-area ad hoc networks,
while the second part proposes new methodologies for evaluation of wireless ad hoc routing protocols.
The educational plan includes developing wireless networking courses coupled with a laboratory, in
addition to proposing novel methods to improve student participation in the learning process.
Ad hoc networks are evolving as very interesting architectures to support pervasive mobile wireless
devices. At the same time, group communication represents a very important class of applications in
future networks. Multicast is the enabling technology for efficient group communication. Providing a
scalable architecture for multicast services in large-scale networks has proven to be a challenge for years
in the networking community. The research challenges are even greater for ad hoc networks, mainly due
to the highly dynamic nature of wireless nodes and their unexpected mobility . The developed multicast
service should allow group participants to join and leave at will, and should impose no restrictions on
node mobility. It should also provide automatic multicast address allocation and session advertisement.
Most existing work on ad hoc networks ignores such requirement. We propose to design a new scalable
architecture for multicast services support in large-scale ad hoc networks, the first of its kind.
Since ad hoc networks are infrastructure-less , the developed protocols must be self-configuring . Scalability and robustness should be addressed carefully with growth in the size of the network and the
number of group participants. Existing approaches for multicast, usually employ flooding or cores for
discovery of group participants. These approaches, however, only apply to small networks due to cost of
flooding and maintenance of the core tree. Questions are often asked about which routing mechanisms to
use, pro-active or re-active routing? Hierarchical or flat? It is usually the case, however, that each of these
mechanisms has its strengths and weaknesses depending on network conditions. We introduce an
approach to use hybrid schemes that adapt to network dynamics. But how to adapt in a meaningful way?
Such adaptation should be a function of mobility and power . The proposed research program intends to fully utilize highly adaptive mechanisms in the various
components of the architecture. Specifically, we design an adaptive zone-based hierarchy augmented by
the notion of contact nodes to increase network coverage. In addition, we provide novel schemes for
multicast service support, based on adaptive anycast for resource discovery. We also introduce
geographic multicast address allocation to map groups into rendezvous regions. All the proposed
mechanisms are self-configurable, and promise desirable scaling and robustness properties.
Once the mechanisms are developed, they shall be thoroughly evaluated using a rich set of test
suites. We intend to use systematic formal methods, network simulation and laboratory experiments to
test, analyze and evaluate the developed mechanisms. In addition, we plan to develop rich scenario-based
mobility models to deeply understand the effect of mobility on ad hoc routing performance.
The above research plan is complemented by a detailed educational plan , potentially leading to
significant improvements in learning . The plan includes development of a series of courses, by the PI, on
protocol design and ad hoc networking and the establishment of a wireless networking laboratory, to
improve theoretical and practical aspects of the networking curriculum at USC. In addition, the PI
introduces a novel and unique approach to increase student involvement in classroom activities using what
we call student-assisted grading and course development. These methods will help create a tighter
feedback loop between instructors and students. We also plan to incorporate architectures and results of
our research into the curriculum for the networking program at USC, at the undergraduate and graduate
levels .
Ahmed Helmy – USC – July 2001 (All rights reserved) 2 1. Introduction
Emerging ad hoc networks are expected to have a significant impact on service paradigms in various
vital fields; such as military, disaster relief, and bio-sensing. These infrastructure-less networks consist
mostly of heterogeneous wireless mobile devices with various power constraints, capabilities and mobility
characteristics. An essential capability in future ad hoc networks is the ability to provide scalable group
services and communication. This proposal presents a novel adaptive architecture to support multicast
services in large-scale ad hoc networks. Unlike existing work on multicast in ad hoc networks that mainly
addresses local areas with tens to hundreds of nodes, our work targets large-scale wide-area ad hoc
networks with, perhaps, tens of thousands of nodes. Issues of scalability, robustness and efficiency in such
highly dynamic environment affect the essence of our architectural design and guide our choice of
multicast model. In order to address these issues, we divide our research proposal into two main areas.
The first is the architectural design and the second is the evaluation methodology. We introduce our
architectural design by discussing the design requirements and presenting overview of the architectural
components. Evaluation methodology is discussed later in the proposal.
1.1 Design Requirements
The main factors driving our design are scalability, the multicast service support, and robustness.
(a) Scalability : Unlike most related work that considers tens to hundreds of mobile nodes, our
architecture should be able to support two to three orders of magnitude more nodes. We believe that
mobile nodes will be pervasive, replacing PCs and cellular phones, with tens of new classes of
application supported by mobile wireless devices (e.g., navigation, location-based services). Flat
architectures are known not to scale well, mainly due to the far-reaching effects of network dynamics;
mobility, failures and topological changes. Such effects consume network resources (i.e., bandwidth
and power), and lead to recovery delays and increased route oscillations. Hierarchical architectures, on
the other hand, alleviate the above problems, as they tend to localize and dampen network dynamics,
and scale routing tables using aggregation. Many existing hierarchical architectures are based on
clustering mechanisms, in which a single node per cluster (called master, cluster-head or parent) is
chosen to manage or organize the cluster. Such architectures suffer from single point of failures, in
which the failure (or movement) of the master may have severe negative effects on the hierarchy.
Furthermore, establishment and maintenance of clusters entails electing the master and mechanisms
for joining/leaving clusters, which usually incur a lot of overhead and complexity. We design an
architecture that leverages hierarchical advantages while alleviating effects of master and hierarchy
maintenance. We adopt a two-level distributed hierarchical architecture. For the first level of the
hierarchy we adopt a zone-based approach (a variant of the zone routing protocol ZRP [28] ), in which
each node has its own view of a zone. For the second level, we introduce a novel concept of contacts
(based on the concept of small-world [29] [55] ) to enhance a node’s view, and aid in route and resource
discovery. Also affecting scalability, is the choice of routing protocol. In general, ad hoc routing
protocols are either pro-active (i.e., table driven) or re-active (on-demand). Pro-active protocols
exchange periodic messages to keep routes up-to-date. Pro-active route discovery has low delay, with
significant overhead of periodic route exchange (many of which may become invalid due to mobility).
Re-active protocols, by contrast, maintain routes on-demand. Re-active route discovery does not incur
periodic overhead, but incurs more route discovery delay, which usually involves request broadcasts
throughout the network. We believe that neither protocol perform well in all network conditions. We
attempt to combine the strengths of both protocols using a hybrid approach.
(b) Multicast Service Support : Multicast participants should be able to join, leave or send packets to the
group at will. Pariticipants are not known a priori and are allowed to move freely during a multicast
session. We design a novel adaptive anycast architecture for resource discovery to facilitate the
rendezvous of group participants. We also introduce a new multicast paradigm instead of the
traditional approaches of broadcast-and-prune or rendezvous cores. Our paradigm is based on sender
push, server cache, receiver pull , approach, that better fits large-scale ad hoc networks. Also, multi-
Ahmed Helmy – USC – July 2001 (All rights reserved) 3 sender groups should be supported. Global multicast groups as well as locally-scoped groups should
be supported. Mechanisms providing participants with active session information should be provided
(This is part of our bootstrap architecture) and a multicast address allocation scheme should be
introduced (We provide a new approach for geographic multicast address allocation). No unicast
routing is required for multicast support. Also, minimum configuration of nodes is desirable (In our
scheme, nodes only need to know a well-known session announcement group address and an
algorithmic mapping function to map groups into regions).
(c) Robustness : In a highly dynamic environment, such as ad hoc networks, where mobility and crashes
are likely to occur, robustness is of prime concern. Being able to adapt to network dynamics to
achieve correct behavior and reasonable performance plays a major role in our design. We incorporate
mobility and stability models into our hierarchy formation to achieve adaptivity. In addition, our
distributed adaptive resource discovery architecture avoids single point of failure scenarios and
promises continued operation and graceful recovery during network partitions. We also incorporate
path redundancy mechanisms in our multicast routing protocol. Multicast trees do not provide
sufficient robustness against mobility and failures. We use mesh structures for robust delivery. Unlike
existing proposals for mesh construction, however, our mechanisms will be designed to build meshes
in anticipation of movement. Not only does that achieve better performance, but also provides path
redundancy that may be used in case of failures. In addition, mesh branches are activated on-demand
to reduce overhead without affecting robustness.
In addition, our mechanisms should be energy-efficient and self-configuring. Energy-efficiency is one
side-effect of scalability. In addition, we attempt to limit communication (one main source of energy
consumption) by using localized mechanisms for advertisement and query. In addition, in resource-
discovery and anycast (where only a single resource is sought), localized broadcast techniques are adopted
to reduce overhead. Self-configurability is an inherent feature of our hierarchy formation and resource
discovery mechanisms.
1.2. Brief Architectural Overview
In order to address the ab ove challenging requirements and support the proposed mechanisms, we
provide an architectural framework based on the following components:
i. Self-configuring hierarchy formation and adaptation
ii. The multicast service architecture consisting of: (a) the multicast model, (b) multicast routing,
and (c) adaptive resource discovery and multicast address allocation.
The rest of this proposal is outlined as follows. Section 2 introduces the hierarchy formation and
adaptation mechanisms. Section 3 proposes a new multicast services architecture including a new
multicast paradigm, multicast routing protocols, and novel architectures for adaptive anycast and
geographical multicast address allocation. Section 4 presents the evaluation and analysis methodology
based on a new scenario-based mobility model. Section 5 discusses related work. Section 6 provides the
proposed research, and Section 7 provides the educational plan.
2. Self-Configuring Hierarchy Formation and Adaptation
We first provide the mechanisms for the self-configuring hierarchy formation, stemming from the
concept of zone-based routing, augmented by our novel idea of contacts , along with description of contact
selection and update mechanisms. Then we propose to integrate mobility adaptation mechanisms into our
hierarchical architecture based on link availability and mobility estimation models.
2.1 Hierarchy Formation
As was discussed earlier, a flat routing architecture does not scale well for wide-area networks, and
more so for ad hoc networks. Also, a hierarchical approach based on concept of master (or cluster-head ),
through which traffic from the cluster funnels, is undesirable.
Ahmed Helmy – USC – July 2001 (All rights reserved) 4 One possible approach to consider is to use the master for infrequent coordination but not for
forwarding packets. One such approach is the Landmark hierarchy (LMH) [32] . Although designed mainly
for wired networks, LMH has the ability to self-configure dynamically, without relying on administrative
domains and exhibits path lengths and routing table sizes comparable to conventional cluster-based
hierarchies (e.g., Internet). The traffic from/to a cluster need not go through the landmark, which adds
robustness. LMH, as presented in [32] , however, was designed mainly for wired networks, without
accounting for mobility dynamics or power constraints. After examination we identified several
drawbacks of LMH. LMH employs complex promotion, demotion, and adoption operations for hierarchy
maintenance. Furthermore, effects of mobility on the hierarchy were found to be drastic, sometimes
leading to total re-configuration of the hierarchy. For example, movement of a high level landmark
triggers re-election in its old region, then demotion and adoption for the moving node are triggered
(potentially for as many times as the levels of the hierarchy). This may trigger a chain reaction that
consumes many resources unnecessarily. In addition, sub-optimal paths may be common due to
hierarchical routing. A variant of LMH (called LANMAR) was presented in [38] . LANMAR, however,
was based on the premise that group mobility will be dominant in ad hoc networks. Hence, it does not
provide mechanisms to solve LMH drawbacks
1 . Another approach that avoids complex coordination for architectural setup is the zone routing protocol
(ZRP) [28] which defines a zone for every node as the number of nodes reachable within a radius of R hops away, shown in Figure 1 (a). Inside the zone proactive (intra-zone) routing is used, so nodes obtain
routes to all nodes within their zone. To discover nodes outside of the zone reactive (inter-zone) routing is
performed by flooding through periphery or border nodes of each zone (known as bordercasting ). ZRP
routing overhead depends heavily on the choice of one parameter; the zone radius . If the radius is too
small the routing overhead is dominated by reactive overhead, and vice versa. Optimizing such a
parameter is nontrivial. A hybrid min-search and traffic adaptive approach is used in [52] but requires
relative stability of the network to approach optimality.
ZRP seems appealing, but experiences excessive delays and overheads in large-scale networks, where
much of the traffic maybe destined out-of-zone. Therefore, we develop a novel approach that goes beyond
the zone while maintaining similar simplicity and stability. Our approach is based on a concept we call
contacts. Contacts, of a certain node x , are nodes that previously existed in x’s zone but are drifting out of
zone, and hence have a network view beyond that of x or any of its border nodes
2 . While drifting away
gradually, x may maintain route to (some of) these drifting nodes using low overhead, by virtue of having
been in x’ s zone and are close to its border nodes. Figure 1 shows a simple illustrative example of zoning,
contacts and effects of mobility.
The concept of contacts relates to the concept of small world graphs [29] . Watts observes that small
world graphs can have low average path length with high clustering. In other words, introducing a small
number of far away links (i.e., short cuts) leads to significant reduction in path length. He also found that
picking random far away nodes may lead to small world graphs, depending on the probability of choosing
the far link. We believe, however, that picking contacts at random is undesirable, since this may lead to
unpredictable overhead for contact route discovery and maintenance. In addition, knowing mobility
characteristics and stability of a node helps identify better (more useful) contacts. We take advantage of
node mobility and pick contacts from those nodes drifting away from the zone. The eventual
characteristics of the formed graphs, however, still need to be studied. The resulting graph is a function of
time and depends on the initial choice of contacts and the node mobility. We plan to study these
characteristics extensively in our research plan.

1 More discussion of hierarchical architectures is given in the related work section.
2 It seems this is one of the few concepts that actually takes advantage of mobility. Mobility is often viewed as a
disadvantage, and for good reasons, but we think it may also be taken advantage of, when possible.
Ahmed Helmy – USC – July 2001 (All rights reserved) 5 Border Node
Center Node
Internal Node
Mobility
R C 5 1 4 3 6 7 2 Zone Radius
Route
R R R R C 5 1 4 3 6 7 2 contact
contact
contact
R R R R C 5 1 4 3 6 7 2 contact
contact
(a) (b) (c)
Figure 1 . Example of zoning, contacts and effect of mobility: (a) Zone for center node C is shown (with radius R ).
Border nodes are numbered (1-7). Nodes 1,3 and 6 are moving/drifting out of zone. (b) Radii for the drifting nodes
are shown. C stays in contact with the drifting nodes, which enables it to obtain better network coverage with low
overhead. (c) After moving away, contact nodes drift up to a point where their zones no longer intersect with C ’s
zone. In this example, C maintains contact with those nodes not more than (2 R +1) hops away, i.e. nodes 3 and 6, and
loses contact with node 1 as it drifts farther than the contact zone.
Contacts play an important role in route and resource discovery as will be explained later. A node
should choose its contacts carefully to attempt to maintain a relatively long and useful contact as long as
the contact route is kept. The contact list , maintained by a node, changes adaptively as the network
conditions change. We further discuss how is the contact list chosen and changed in the next section.
2.2 Hierarchy Adaptation
The architecture presented thus far provides a framework for hierarchy formation. Such hierarchy
needs to be highly adaptive to network dynamics and conditions. Unlike traditional protocols, our
protocols must integrate concepts of mobility and energy . To achieve this, mechanisms that integrate
measures of stability and power consumption must be introduced. We use such mechanisms for hierarchy
formation and route selection. In hierarchy formation, the zone radius and contact selection may be
adapted dynamically to establish certain stability measures for zones and contacts. Furthermore, each node
should adapt its behavior to achieve desirable collective performance. For example, number of contacts
chosen by a node should be a function of the contacts already established by other nodes in its zone
Increase in number of contacts increases bandwidth and must be done only as necessary.
In [13] a scheme was proposed to provide mobility-based adaptive behavior, where a model is used to
measure link and path stability . The model used is ( a,t) , where a is the probability that the link will be
available for time t , and was utilized to build adaptive clustering schemes for ad hoc networks. The basic
idea is to build stable clusters in which internal route availability is probabilistically bounded. The
proposed model is a random walk-based model that determines the conditional probability that the nodes
will be within range of each other at time t 0 +t given that they are located within range at time t 0 . Although
quite useful for analytical reasoning about mobility, the random walk mobility assumption may not hold
for correlated mobility. Potential benefits of this approach warrant further study and investigation for
different mobility models. Another work [57] defines a mobility metric derived from signal strengths,
using a free space propagation model;
d TxPwr
RxPwr
2 1 a
where RxPwr ( TxPwr ) is received (transmitted) signal
power, and d is the distance between transmitter and receiver. By measuring the signal strength between
two consecutive packets received from the same transmitter a relative mobility metric is defined as
Ahmed Helmy – USC – July 2001 (All rights reserved) 6 old
new
RxPwr
RxPwr
log 10 , if negative high value then nodes are moving away quickly, and vice versa. This metric
was presented to aid in choosing cluster-heads. We plan to study such concepts, analyze their limitations,
suggest improvements and extend them to use with other criteria such as power and QoS measures [13] , and integrate it with our hierarchy formation architecture. Such integration helps in contact selection,
where a node chooses its contacts with probability p , as a function of the border’s mobility, the number of
zone contacts, energy and activity. One simple model to be initially used is to have p proportional to the
energy estimates of both nodes E est
, their relative stability S est
, the activity level of the node A est
measured
as rate of discovery requests. Also, p is inversely proportional to the number of zone contacts Z est
. Hence,
est
est est est
Z A S E p a
. These measures are based on either local or zone-scoped (e.g., piggybacked on intra-zone
pro-active messages). E est
includes energy estimates at the node choosing the contact, and the contact node
drifting out of zone . To accommodate heterogeneous nodes, the estimates should include energy left E left
, and the drainage D E . Thus, a simple equation for the energy estimate is:
contact
left
node
left
est
E E E E E ⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
Δ
⎟
⎟
⎠
⎞
⎜
⎜
⎝
⎛
Δ
=
. Stability estimates will be derived from modified models based on the ( α , t ) model and the signal
power mobility metric described above, subject to further research. Furthermore, as was pointed earlier,
the choice of contacts (along with network dynamics) affects the resulting connectivity graph. Thus, we
consider such choice to be non-trivial, especially if we want to attain small world characteristics. We plan
to pursue this issue further in our research. These adaptive mechanisms are also used in the multicast
service model and resource discovery in section 3. In such case, contact’s choice should also be affected
by the capabilities of the contact node (e.g., GPS capability, or being a session or sender discovery server).
3. Multicast Service Architecture
Providing a scalable multicast service architecture is the focal point of our proposal. The
hierarchical architecture proposed thus far provides the basis for efficient support of our multicast service
model, presented in this section. As the mechanisms for our multicast model unfold, the essential role
played by the adaptive hierarchical architecture will become very clear in support of multicast routing and
resource discovery. In general, existing work on ad hoc multicast concentrates on multicast routing in
small to medium size networks. Particularly focusing on mechanisms to establish distribution trees (or
meshes) between senders and receivers, mainly using periodic broadcasts (either from the sender [36] or
receiver [39] side) or relying on the existence of cores [26] . In this proposal, however, we address multicast
in large-scale ad hoc networks, and propose new schemes for bootstrapping multicast services, by
providing efficient resource discovery using popularity-based adaptive anycast and geographic multicast
address allocation . We are not aware of existing or on-going work on multicast for wide-area ad hoc
networks. Furthermore, we do not know of other work for bootstrapping such service, resource discovery
or multicast address allocation in ad hoc networks. Our work does not assume any underlying unicast
routing protocol.
Our multicast service architecture consists of three main components; (a) the multicast model (i.e.,
how senders and receivers meet), (b) multicast routing (i.e., establishing multicast distribution paths), and
(c) adaptive resource discovery architecture using anycast and geographic multicast address allocation.
3.1. The Multicast Model
The basic premise for scalable multicast is that sources do not know who/where receivers are a
priori. This model enables any node to join or leave a multicast group at any point in time. Hence, one of
the main components of multicast is a mechanism for group participants to meet or rendezvous.
Traditionally, this problem has been addressed, in wired and ad hoc networks, using two different
approaches; broadcast-and-prune , and rendezvous cores . In the former [15] [42] [36] , a participant (usually
the sender) announces its presence by broadcasting data packets (or control messages) throughout the
network. Network nodes not interested in the group send prune messages to stop the flow of packets (or
Ahmed Helmy – USC – July 2001 (All rights reserved) 7 simply do not respond in case of control messages broadcast). These broadcasts are periodic to capture
network and membership dynamics. It has been shown that such model is best suited for small to medium
size networks with densely populated groups, but does not scale for wide-area networks [60] . The
rendezvous cores approach, by contrast, avoids periodic broadcasts by providing explicit join mechanisms.
Participants join (or send packets) towards a common core, which relays the packets from the senders to
the receivers using a shared tree (or mesh) [2] [26] . Problems of traffic concentration or single point of
failure scenarios for core-based approaches may be alleviated by using multiple cores or dynamic core
election mechanisms. We believe, however, that the major research problem associated with the core
approach is the core bootstrap and consistency problem. How do participants know the core’s
address/location? Senders and receivers need to maintain a consistent view of the cores in order to meet.
This problem was addressed for wired networks [14] within a single domain, and uses a flooding to
disseminate core-to-group mapping. This scheme does not scale well for wide-area networks and its
convergence performance degrades with the size of the network. The problems of rendezvous cores are
exacerbated by ad hoc network dynamics and mobility.
Unlike the previous approaches for multicast, we instead propose a new multicast model. We refer
to our model simply as sender push, server cache, receiver pull model. Unlike previous work on ad hoc
multicast that requires periodic broadcasts throughout the entire network, our scheme incurs less overhead,
and only when necessary as necessary and as localized as possible. We introduce the notion of sender
discovery servers (SDS) to aid in sender location and information dissemination. As shown in Figure 2 , a
sender sends an Advertisement ( Adv ) using localized broadcast . SDS s receiving the Adv store this
information. Receivers send joins toward the sender based on backward learning ; every node forwarding
the Adv adds its address to the message to construct a path back to the source. Other, farther, receivers not
receiving the Adv message attempt to find a nearby SDS , first by checking in their own zone ( SDS s advertise in their zone), then by checking with their contact list. If SDS is found it is queried for group
information and responds with a join reply, with approximate source location or possible routes (if
available at the SDS ). Depending on the quality of the provided routes (if any), the querying receiver(s)
may opt to use these routes or use zone/contact search for other routes (eventually, geographic routing
may also be used for route discovery as in LAR [35] , as described later). If SDS is not found, a receiver
may send a localized broadcast to discover other nearby receivers of the group. In case this process fails to
get information about the group then a fallback mechanism, described later in this section, is used. Once
the information about the group/senders is available the route discovery/construction is initiated. An
illustrative example is shown in Figure 2 . R 3 SDS 2 R 4 R 1 S SDS 1 R 2 Sender Adv.
Join Query
Join Reply
Join Request

R 3 R 4 R 1 S R 2 (a) (b)
Figure 2 . Multicast service scenario. (a) Sender S becomes active and broadcasts an advertisement (Adv.) locally
(covered region is shaded). Sender discovery servers SDS
1 , SDS
2 , and joined receiver R 1
receive the Adv. When new
receivers join the group, they try to find: (i) a sender discovery server, (ii) nearby members, and/or (iii) active
senders for the group. Receiver R 2 finds R 1 using local broadcast, while R 3 and R 4 find SDS
1 and SDS
2 . (b) Once
sender information is obtained, receivers send Join Requests to establish multicast distribution paths.
Ahmed Helmy – USC – July 2001 (All rights reserved) 8 3.2 Multicast Routing
Establishing multicast distribution paths for ad hoc networks has been shown to be more robust using
mesh structures, as opposed to conventional tree structures [27] . For single-source groups or when sources
are sparsely distributed, however, even a mesh does not provide the desired path redundancy. Local
recovery mechanisms [56] may be used to alleviate such problem. Conventional receiver-initiated
multicast schemes setup reverse path forwarding (RPF) trees [2] due to dependence on unicast routing. For
ad hoc networks, however, using RPF paths is not desirable due to possible path asymmetry due to
wireless channels. Furthermore, we do not use RPF nor rely on existence of a unicast routing protocol.
We propose to use mesh structures with local recovery mechanisms. In addition, multiple paths
may be selected by the receiver (during the discovery process) to increase robustness of the multicast
distribution mesh. Depending on the number of senders in the group and their location (if available), in
addition to the mobility of the receiver, a receiver may opt to choose several stable paths to join the group
(when propagated, route information includes stability metrics). The receiver sets an ‘active’ flag in only
one join to activate only one path at a time. Only active paths forward data packets. This reduces packet
transmission overhead (a very significant factor in ad hoc networks) while maintaining robustness. If
performance or stability of the active path degrades, or local recovery fails, the receiver may activate
another path with high stability. Also, when the receiver moves, it may activate another path containing
one of the new neighbors, thus achieving fast handoff using the concept of multicast-based mobility [1] . Our investigations show that, on average, a moving node traverses 2.5 hops to reach the nearest point of
the multicast distribution structure in very large networks (with up to 5000 nodes)
3 . Moreover, mobility
prediction [37] may be used to achieve advance joining, further reducing the effects of mobility handoff
(i.e., mobility of the receiver in ad hoc networks). Rules for activating/de-activating should be carefully
selected to avoid black holes. For example, if any member exists downsrtream a branch then the branch
must be activated. In order for a branch to be inactive, all downstream branches must be inactive.
So far, we have assumed that the receivers’ efforts in searching for group information are successful,
using the above localized search techniques. In case of sparse groups, where participants are far apart, or
in case of multi-sender groups, where information obtained from other group members may not be
complete, a fallback mechanism should be used. Such mechanism should be efficient, avoiding frequent
global flooding and should be adaptive to membership dynamics. To achieve this, we introduce a novel
bootstrapping anycast architecture for multicast service in ad hoc networks, discussed next.
3.3. Adaptive Resource Discovery and Multicast Address Allocation
The major research challenge for multicast resource discovery (i.e., discovery of g roup address,
senders), is the lack of any (centralized or distributed) infrastructure to hold and distribute such
information. Inter-domain multicast for wired networks [45] [3] utilizes the AS hierarchy of the Internet
and uses BGP extensions to distribute multicast routes that map group prefixes into root-domains
(established by the multicast address allocation). In turn, receivers join towards the root-domain and
senders send their packets towards it. Intra-domain multicast is used within the domains and is built on top
of unicast routing. All these infrastructures (unicast routing, AS hierarchy) do not exist in ad hoc
networks.
An architecture for anycast routing in the Internet was recently proposed in [31] . Utilizing hierarchical
routing and aggregation. This work provides a scalable mechanism to discover members of anycast groups
that are closer to the requester than other members. The architecture identifies two mechanisms, a low
overhead mechanism, using default routes, for non-popular anycast groups (in which case the request is
routed to the home domain, derived from the anycast address itself), and another mechanism for popular
anycast groups that caches routes for nearby members.

3 These results were obtained for Internet topologies. We shall investigate these measures further in the context of ad
hoc networks to study how they differ.
Ahmed Helmy – USC – July 2001 (All rights reserved) 9 The only global infrastructure we can probably utilize in ad hoc networks is geographic location . Based on geographic multicast address allocation, we devise a new adaptive anycast architecture as
follows. The multicast address space is broken into prefixes. Each multicast address prefix is assigned to a
geographic region called the rendezvous region (RR)
4 . Nodes located in the RR have a collective
responsibility of maintaining information about the groups belonging to the group prefix assigned to their
current region. Since it is ‘collective’ and could be done by only a small subset of SDS nodes (say 3-7
uncorrelated nodes) we can use a probabilistic promotion scheme. Each node decides locally whether it
will become a SDS based on its own configuration (some nodes maybe configured as servers), capabilities
(e.g., GPS), power and stability estimates. If so, it obtains its (approximate) geographical location
5 and
determines the group prefix to which its current location maps, using algorithmic mapping in the general
form of f(x
1 to x
2 , y
1 to y
2 ) = G
prefix
, or similar
6 . At that point, the node acts as a member of the anycast
group of SDS s responsible for G prefix
, and advertises this information in its zone and to its contacts
7 . Other
SDS s for the same prefix reply to update the new SDS (the reply is localized to reduce overhead). As SDS
nodes move out of the RR for the corresponding G prefix
, they advertise their latest group information and
leave notice to the RR (using geocast [43] , for example), which increases the probability of other nodes
promoting themselves to become SDS s. This ensures constant replenishing of the pool of SDS s serving as
members of the anycast group for that RR . The above scheme requires approximate knowledge of geographic location. We do not assume that all
nodes are GPS capable. We do assume, however, that nodes are heterogeneous, and that some nodes will
be more capable than others, specifically with localization capability
8 . We shall investigate GPS-less
techniques [58] [59] in our work.
Session Initiation A node initiating a multicast session is expected (without necessity) to use the
above algorithmic mapping to obtain a multicast address that maps into a geographical vicinity as its RR . In any case, group/session initiation requests/updates are sent to the RR to avoid collisions in multicast
address allocation. When a new sender of a group belonging to G prefix
becomes active it performs a
localized broadcast (as described earlier) and, if far from its RR , issues an update to RR. The requests and
updates are sent using lollipop-LAR (our modified location aided routing (LAR) [35] ) to improve
scalability. In lollipop-LAR, a far away sender chooses a contact that is closer to the RR . If the distance
between the contact and the RR is less than a limit l , the contact sends the request/update to the RR directly
using LAR, otherwise it chooses one of its contacts closer to the RR . Estimating l is part of our research.
When a node joins the group, it first attempts localized search for SDS . If this fails it sends/geocasts a
join query to the RR using lollipop-LAR. A join reply is issued by a SDS in the RR and follows the
backward route created by the join query (this path is only used for the reply so it need not be the shortest
path). Once the receiver has group/sender information it sends a Join request as described earlier to
establish multicast distribution path(s). Join latency is subject to further study.
Popularity-based Dynamic Adaptation If used for the common case, geocasting to the RR may
incur a lot of overhead to maintain group information, especially for a very dynamic environment as is ad

4 This does not necessarily imply, however, that these groups are geographically limited to that region.
5 Geographic location update need not be done frequently, only when a node moves noticeable distances. Afterall,
this information is approximate and is used for distributed resource discovery (not for forwarding packets).
6 We assume that nodes wishing to use multicast will express their desire to use multicast to their neighbors, zone or
contacts (similar to a subscription) , which will result in a reply with the well-known algorithmic mapping function.
The scheme should allow for changing this mapping to another well-known mapping (very infrequently, though).
The choice of rendezvous region and the algorithmic mapping function will be subject to our further investigation.
7 Alternatively, a node may advertise this information using localized broadcast, or geographically scoped broadcast
(where only nodes, within a specific region, broadcast the packets), or Geocast [43] . We shall investigate these
alternatives in our research.
8 This poses questions about accuracy of location estimation and its effect on performance of the proposed
mechanisms. This is an important issue to study through investigation of space of possibilities of number of
localization capable nodes, and their distribution (dense/sparse), under a wide array of mobility patterns. We shall
address this issue in our proposed research.
Ahmed Helmy – USC – July 2001 (All rights reserved) 10 hoc networks. Hence, our multicast service and resource discovery paradigm should adapt to dynamics of
membership, to achieve better performance for popular groups. As explained above, sending or joining
groups entails local advertisement (Adv) or query. This gives indications of the popularity of the group in
the vicinity of the participants. Nodes receiving Advs and queries, those that are willing to become SDS
for that group (based on their configuration, stability and capability), estimate the popularity of the group.
The initial estimate is based on Advs/queries heard. If this initial estimate exceeds a threshold pop query-th
a
local group query is sent by the candidate- SDS to its own zone (using pro-active routing updates readily
sent) and to its contacts. This group query gives a better estimate of group popularity Grp
est
, in addition to
information about existing SDS s nearby SDS
est
. Popularity estimate pop est
is be obtained as
est
est
est
SDS
Grp
pop a
. If
pop est
> pop
th
where pop th
is the popularity threshold (subject to further research), then a node advertises
itself as SDS for the group to its zone and contacts, and contacts the RR SDS for updates (using lollipop-
LAR). Future nearby join queries for the group reach the local SDS and are answered locally, reducing
overhead and delay. Furthermore, this adaptive mechanism also achieves better robustness and continued
operation during network partitions.
Discussion We note that the probability of success of the localized search is affected mainly by two
factors. The first is the is the group address, obtained during session initiation, which decides the location
of the RR , and in turn determines the location of the RR-SDSs . The second factor is the nearby popularity
of the group nearby, which decides the promotion of nearby SDSs. Many of the offered services are
expected to be location-based services, meaning it is targeted to a specific location. Hence, these groups
will tend to be popular within certain locations (more than others). In addition, initiators are expected to
choose group addresses that have RR in the geographical vicinity. Both these factors increase the
probability of success for localized search, and lead us to believe that, in the common case, our
architecture is capable of high performance, with low overhead and low delays. Our schemes work for
global as well as locally-scoped groups.
Several questions arise for the above proposed architectur e. Some of which we try to address here. For
example, how do participants know about newly initiated sessions and their properties? This can be
provided using the same scheme provided above, as follows. When multicast participants express their
interest in multicast service they obtain the algorithmic mapping function (described above) as well as a
well-known session advertisement group address. This address also has RR . As groups are initiated, they
are updated at the RR-SDS and the local SDSs (if any), then information about new sessions is obtained as
above. This provides an bootstrapping mechanism essential for multicast service.
Other questions include the frequency of update to the SDSs, and refresh of multicast state using soft-
state. Also, what about multicast state kept in the nodes? How does such state scale with the number of
active groups and sources? What if for many applications statistical information is sent periodically, but
infrequently, to the whole group (as in RTCP). Should these infrequent senders be treated as regular
senders or should they use a shared tree? What are the effects of node failures, network partitions and
different mobility patterns on the performance? These issues are subject to further research.
4. Modeling, Analysis and Evaluation Methodology
Unlike traditional networks, ad hoc networks tend to be highly dynamic, mainly due to mobility.
Mobility affects protocol behavior and performance directly and drastically. Some even suggest that many
results obtained for ad hoc network studies merely reflect the mobility model used [61] . We believe that it
is imperative to study and understand the effects of mobility on the performance and correctness of our
proposed mechanisms. Specifically, we follow an approach that allows us to systematically relate certain
mobility characteristics (e.g., movement correlation) on our mechanistic building blocks.
In general, previous studies on ad hoc networks used a single mobility model, usually the random
way-point model. Such model fails to capture mobility correlation that may exist in real world scenarios.
Instead, we propose a novel, yet simple, approach based on scenarios or state machines. In a specific state
(e.g., pedestrian or vehicular) the node has a certain mobility model. Upon change of state (e.g., bus) the
characteristics of the model change. Sequence of states is derived from real-life scenarios. Such model
Ahmed Helmy – USC – July 2001 (All rights reserved) 11 includes richer mobility semantics; such as group mobility (e.g., bus), correlated mobility (e.g., vehicles
on a congested highway), or random movement (e.g., pedestrian).
Often, network simulation is used to evaluate ad hoc routing protocols. Choosing a meaningful set of
simulation and testing scenarios, however, is often hard. The STRESS methodology [5] attempts to
alleviate such problem via test scenario synthesis.
We plan to use three different approaches for evaluation; systematic algorithms for robustness
validation using STRESS , packet level simulation using the network simulator (NS-2) [4] , and
implementation in a laboratory setting. The tools shall be adapted adequately for our studies. We focus on
analysis and understanding of mobility effects, rather than attempting to develop an accurate mobility
model based on actual traces.
STRESS The STRESS methodology [5] (designed and developed by the PI) provides algorithms for
systematic and automatic evaluation of networking protocols. Such methodology targets multiparty
protocols , in that it is different than traditional validation techniques. To capture the nature of multicast
protocols, STRESS provides a model for the network topology and the states of the different group
members using a global finite state machine (GFSM) model. In addition, network fault models (such as
packet losses, or machine crashes) are incorporated into the algorithms to facilitate robustness checking.
In addition, this method introduced the concept of topology synthesis that enables backward search; found
in our case studies to be much more efficient than traditional forward search. Hence, we believe that
STRESS includes rich models and semantics needed for systematic evaluation and robustness verification
of our proposed protocols. We are not aware of other work with these characteristics.
STRESS provides two main algorithms to search the protocol state space, fault-independent test
generation (FITG) based on forward search [46] , and fault-oriented test generation (FOTG) based on
topology synthesis and backward search [6] . Case studies have been conducted using this methodology for
analysis of multicast routing protocols and reliable multicast protocols, among others. However, STRESS
has only been applied to LANs and end-to-end models (using a virtual LAN model). It has not been
applied to multi-hop models. It also does not include rich mobility models. This entails changing the
connectivity of a node (or several nodes to account for correlation). These mobility models need to be
embedded into the methodology. The scenario-based approach we take for mobility modeling lends itself
readily to a finite state machine (FSM) representation. Whereas correlation between node movements may
be captured using the global FSM. We shall also investigate other aspects of mobility (such as velocity
and direction) and location, in the context of STRESS. [We do not elaborate for brevity.]
Scenarios resulting from the stress studies will carry on to the simulation stage.
Simulation The network simulator (NS-2) [4] (developed partially by the PI) supports a variety of
multicast and ad hoc routing protocols, and will be used for simulations. Several evaluation metrics shall
be used to evaluate the performance of the proposed architecture, including packet delivery ratio, join
latency, route discovery delay, and control overhead, among others. The salient feature of our work,
however, is the integration of physical constraints into the topology map. This facilitates modeling of
mobility scenarios. For example, overlaying of highways, routes and bus movement enables us to capture
different correlation models. We plan to develop a centralized mobility coordinator object to correlate
node movements.
Implementation We plan to implement the proposed architecture in a laboratory test bed. (The USC EE
networking laboratory is founded and directed by the PI). Using carefully designed experiments, the
abilities of each component and their integration shall be demonstrated. The test bed currently consists of
around 20 nodes (to be extended later) of laptops and iPAQs running Linux and wireless Motes running
TinyOS. Also, the simulations may be run in emulation mode [17] (supported by NS-2) enabling
simulations to interface to an operational network. Earlier developed tests will be leveraged. Details
omitted for brevity.
Ahmed Helmy – USC – July 2001 (All rights reserved) 12 5. Related Work
Related work lies in the areas of ad hoc routing (unicast and multicast), hierarchy and cluster
formation, anycast architectures, inter-domain multicast, geocasting, and mobility modeling. Note that
only closely related work is discussed, given the research interest given to ad hoc networks recently.
In the area of unicast ad-hoc routing, protocols are generally classified as either pro-active (or table-
driven) or re-active (or on-demand) protocols. Pro-active protocols include DSDV [18] , CGSR [19] , and
WRP [20] , and rely upon routing updates to maintain consistency of route information. Re-active protocols
include AODV [21] , DSR [22] , TORA [23] , ABR [24] , and SSA [25] , and create routes only when required
by the source node. One feature of SSA is that it selects routes based on the signal strength between nodes.
Fisheye state routing (FSR) [65] is used to reduce routing update overhead of link state routing. For a node,
the route update frequency to a certain destination is inversely proportional to the distance (in hops) of the
destination. That is, routes to nodes within a small distance are sent to neighbors with higher frequency
than routes to far away nodes. This reduces route overhead (not table size) and reduces accuracy of
routing with distance. Routing efficiency decreases and delays increase, however, with dynamics of
mobility, and routing table size grows linearly with network size [38] . CEDAR [40] uses a simple
approximate algorithm was proposed for building a core graph consisting of the minimum dominating set
of nodes. The mechanism was designed for a network of small (10s of nodes) to medium (100s) size
network. A variant of CEDAR [41] was used for multicast by joining to the core graph. The effects of
mobility and concentration on the core graph, were not clear in the study. Other works on multicast ad-hoc
routing in [26] [27] generally extend existing multicast routing for the Internet, such as PIM-SM [2] . Other
recently proposed multicast ad hoc routing protocols include tree-based and mesh-based protocols. Tree-
based protocols include AMRoute [33] and AMRIS [34] . AMRoute creates a bi-directional shared core-
based tree using unicast tunnels, it uses virtual mesh links for tree creation and needs unicast, but incurs
temporary loops and chooses sub-optimal routes with mobility. AMRIS uses a shared tree and an ID
number per node, does not need unicast, broadcasts new-session messages and uses beacons to detect
disconnection and re-joins to potential parents. However, it uses the expanding ring search mechanism for
branch re-construction due to node failure, which does not scale well. Mesh-based protocols include
ODMRP [36] and CAMP [47] . CAMP uses a shared mesh, and all nodes keep membership, routing and
packet information. New members use expanding ring search to find other member neighbors. However,
CAMP needs a special unicast protocol for its proper operation. ODMRP floods packets within mesh, but
follows an on-demand policy for establishment and update of the mesh. It uses request and reply phases,
broadcasts source announcements, and does not require unicast routing. The mesh is created when join
requests from multiple receivers are sent to multiple-sources. Hence, for sparse groups or single-sender
groups ODMRP may not be as robust. A local route recovery scheme [56] may be used to address this
problem. In our routing protocol, we utilize the concept of mesh construction and local recovery, but we
attempt to avoid floods for resource discovery, using a contact-based query approach. Moreover, we allow
on-demand-activated multiple paths to be constructed to a single source, to increase robustness and
achieve better handoff performance during mobility. Multi-path routing was proposed [49] for parallel data
distribution. We will leverage mechanisms provided for multi-path discovery, but we do use multiple-
paths for multicast differently, by activating on path at a time on-demand.
Recently, a scalable anycast architecture was proposed for the Internet [31] , and was discussed earlier.
Work on location-based routing was presented in location-aided routing LAR [35] , Geocast [43] [44] . We
use concepts of geocasting for route and resource discovery. We modify it for scalability using lollipop
shaped regions with the aid of contacts. We shall also leverage work on GPS-less positioning [58] [59] to
determine relative approximate positions of the nodes, assuming GPS capability in some nodes.
One of the earliest works on self-configuring hierarchical architectures includes work on the landmark
hierarchy Landmark hierarchy uses the concept of area hierarchy. An area is constructed by grouping
nodes into logical areas and subareas. Each node has a level in the hierarchy and a radius r associated with
that level. Each node advertises information about itself to nodes within r hops. So, a node receives
advertisements from nearby nodes that are the lowest level of the hierarchy, and faraway nodes that are at
higher levels of the hierarchy, and so on. In effect, each node knows less information about farther nodes.
Ahmed Helmy – USC – July 2001 (All rights reserved) 13 In [50] landmark hierarchy is used to form an object location architecture for sensor networks. Hierarchy
levels are self-configuring and may be adapted using a promotion/demotion scheme. Drawbacks of
landmark hierarchy were discussed earlier. LANMAR [37] [38] uses the landmark hierarchy concepts to
establish hierarchy in ad hoc networks. However, landmarks are used for sets of nodes moving together as
a group to reduce routing information exchange. Other hierarchical ad hoc routing include the zone-based
hierarchical link state (ZHLS) [51] . ZHLS is a global positioning system (GPS)-based routing protocol for
ad hoc networks, where a network is divided into non-overlapping zones. A node only knows node
connectivity within its zone and the zone connectivity for the network. This architecture does not use
cluster head to mitigate traffic concentration, reduce routing protocol control/message exchange overhead
and avoid single point of failure. It uses zone ID and node ID for routing. However, the zone map is
defined by design for interzone routing, and hence does not adapt to network changes and dynamics.
Another protocol called the zone routing protocol (ZRP) [28] [52] discussed earlier. In [30] the ZRP
approach is coupled with geographic (geodesic) routing for remote routing. In [13] [53] [54] the link
availability model discussed earlier is proposed. The authors suggest to use it with a cluster based
approach, in which a parent is selected based on the availability model to increase the lifetime of the
cluster. Parent selection and cluster dynamics may complicate our architecture. Instead, we propose to
incorporate the availability model with our modified ZRP approach and to use it for determining zone
size, and in choosing contacts.
In the Internet, Hierarchical PIM [45] was proposed is an inter-domain architecture based on the PIM-
SM protocol. It suggests a hierarchy of rendezvous points (equivalent to cluster heads) to communicate
between multicast domains. The BGMP architecture [3] was proposed for hierarchical inter-domain
multicast. It uses a bi-directional shared tree and the notion of a root domain. The problem of multicast
address allocation is coupled with BGMP for the choice of the root domain. The same study proposes the
MASC scheme for multicast address allocation. Such problem is still active in research.
In the area of mobility and movement modeling, several models have been developed for cellular and
ad hoc networks. For cellular networks, the random-walk model was used in [7] [8] [9] , where the next
position is picked randomly then the speed and direction are picked uniformly from a given range. [7]
studied the cell residence time and found the generalized gamma distribution to be suitable, while the
negative exponential distribution was suitable for channel hold time modeling. Rappaport also studied the
cell residence time (also called the dwell time) and used models based on multi-dimensional birth-death
process. [10] used a model based on semi-Markov process to represent mobile user behavior. [11] noted
that mobility patterns tend to increase during rush hours, while [12] used a group mobility model to
evaluate ad hoc routing protocols. [13] used a random-walk and pause-time random models to derive
probability of path availability as function of time, for clustering mechanisms in ad hoc networks. Other
models include pursuit and tracking, nomadic and brownian motion.
These models affect the performance evaluation of our architecture. For example, group mobility adds
correlation that is not exhibited by random moves, which may create traffic or state concentration in the
network. We shall study these models, investigate their effect on performance, and possibly suggest
improvements or modifications as necessary. Our protocols should be evaluated using a variety of
mobility models.
6. Proposed Research
We propose to conduct in-depth studies to understand the architectural design trade-offs of the
proposed protocols. We propose to pursue research in two major directions (1) architectural and protocol
design and (2) systematic evaluation methodology . Particularly, we intend to investigate and attempt to
answer the following questions (in addition to questions discussed earlier):
- How should the list of contacts be chosen? and what is the effect of introducing the contact list on the
overall performance of the architecture? Some performance metrics are given in the evaluation
section. What are the properties of the resulting network, as function of the probability of choosing
contacts p , and how does it change with mobility? We believe that such network may approach a
desirable small-world graph for careful choice of p and its parameters, resulting in reduction of route
Ahmed Helmy – USC – July 2001 (All rights reserved) 14 and resource discovery overhead at the expense of increased latency than pure ZRP in some cases. We
plan to investigate that trade-off in depth. Another (longer-term) issue includes several levels of
contacts. We have only discussed nearby contacts (let us call them level-1 contacts). Would it be
beneficial to include farther contacts in the architecture (call them level-2 or higher)? There is an
interesting trade-off to investigate between increased contacts and increased maintenance overhead.
- What is the effect of using different algorithmic mapping functions, group initiation procedures, and
SDS promotion scheme? We think that with the proper choice of the promotion parameters these
effects will be reduced. How do we choose pop th
to achieve desirable effects? The effect of
membership distribution will also have an effect, we expect. By thorough investigation, using various
membership distribution patterns and densities, we hope to develop enough insight to suggest useful
algorithms, or heuristics, to choose the parameter settings adaptively.
- How do characteristics of the mobility model affect behavior and performance of our protocols? We
plan to investigate such effects thoroughly using a semantically rich scenario-based mobility model
explained in the evaluation section. Our current intuition is that short and long term mobility
characteristics matter. Specifically, we think correlated mobility, affecting node density, will have a
strong effect. But to what extent? By using appropriate mobility and stability metrics we hope to
reduce undesirable effects, where the mechanisms adapt intelligently to network dynamics. We plan to
test our adaptive schemes by embedding mobility models into STRESS and NS-2. We plan to use a
two-level FSM model. A transportation state (e.g., pedestrian, vehicular, train) captures the long-term
characteristics and correlation of movement. While mobility states within a transportation state
represents short-term characteristics (e.g., low or high mobility). Transportation literature is helpful in
this process [63] [64] . We also plan to develop a multi-hop model for STRESS using the notion of node
neighborhood , which allows independent network view for each node.
- How are geographic information detected using partial capability of GPS in conjunction with GPS-
free techniques? Furthermore, how is this information updated to the SDSs by the senders? We plan to
use only approximate estimates of location, to reduce update frequency (as function of the sender
mobility), and use location recovery and location-aided routing techniques to locate senders.
Estimating these trade-offs and finding a reasonable mix of location techniques needs more research.
- What is the effect of failures on the performance and robustness? Power depletion, crashes, partitions
and partition healing are but a few of these failures to study. We plan to fine-tune our adaptive anycast
scheme and multicast routing to increase robustness. The effects of multi-path and channel fading,
however, can only be studied in an implementation environment, and we plan to investigate them in
our laboratory during the later stages of this research.
- How does the multicast state scale in the network nodes? This is an important aspect to analyze as
function of the number of groups and sources. We leverage our previous work in that area [62] . - What is the quality of the routes discovered? We expect sub-optimal routes to result due to the local
recovery mechanisms. We plan to study and analyze such mechanism in detail.
Other problems to investigate include modified or new mobility estimation and prediction schemes that
attempt to avoid limitations of existing methods (such as depending on random models or using simple
free-space model). Also, integrating QoS into these models, (as hinted in [13] ) is to be investigated.
Longer-term research issues to pursue include the use of mobility prediction schemes for fast handoff
using Multicast-based mobility concepts [1] , the study of asymmetry and uni-directional links on route
quality, and possible development of localized shared trees for low-rate, densely populated sources.
7. Networking Educational Plan
This educational plan includes two elements. (1) A series of courses (accompanied by a
networking laboratory) to be developed and instructed by the PI to improve theoretical and practical
aspects of the networking curriculum at USC. (2) A new approach to involve students in various aspects
of the classroom, including course preparation, evaluation and examination.
Ahmed Helmy – USC – July 2001 (All rights reserved) 15 (1) Lab Courses Plan (Practical Networking)
This plan includes a series of networking courses (integrated with the networking lab) both at the
undergraduate and graduate levels, over the next few years.
- EE-555 Broadband Networking Architecture: LANs, WANs, congestion control, queuing disciplines,
traffic characterization, admission control, switching, optical networks, multicast routing, wireless
networking and IP mobility.
- EE-599 Internet Protocol Design, Simulation and Testing: DiffServ and AQM, unicast, multicast,
geocast and hierarchies in ad hoc networks, TCP and reliable multicast over wireless, Stress testing,
multicast congestion control. (Developed and instructed by the PI. For graduate students)
- EE-499 Introductory Computer and Wireless Networking Laboratory: packet, route and topology trace
and measurement tools, route configuration and test bed setup, simulation vs. implementation,
congestion control, mobile IP, wireless ad hoc networks. (Developed and instructed by the PI. For
undergraduate and entry-level graduate students)
- TBD Advanced Wireless Networking Lab: IP mobility with fast handoff, ad hoc networks, sensor
networks.
- TBD Protocol testing, verification and simulation: (with VLSI chip testing)
The PI has founded and is directing the networking lab at the EE department at USC.
(2) Student-Assisted Education and Evaluation
The PI proposes the concept of student-assisted grading and course-development , in which students
grade assignments (as double-blind review), based on adjustment questions and supervision of the
instructor. Students also participate by suggesting experiments, assignments and exam questions. A novel
instant feedback system is proposed for in/during/out-of-class evaluation. This new approach for student
participation gives the students the opportunity to interact, innovate, and review technical material in a
new way, which would enhance their level of understanding.
Student-assisted grading system:
This is a new approach for assignment grading and evaluation, especially in a lab setting, to be
included as one of the experiments in the classes. In order to evaluate experiment assignments, each
student is assigned a number of assignments to grade for classmates (anonymously, using double-
blind review). Before the students start the grading, they need to answer several adjustment questions
9 carefully chosen by the instructor. The answers provided by a student give a weight for the student,
denoting the level of understanding. After the students grade the assignments, the grades provided are
adjusted according to the grader’s weight. The instructor samples these graded assignments to
evaluate the process and its fairness. This will be coupled with an incentive system that rewards good
grading and punishes dishonesty and inconsistency (measured as high variation between adjustment
question performance and grading performance).
This grading experiment provides a chance for the students to view colleagues' work in detail, to
learn and be able to improve and compare their own performance with that of their colleagues. Since
students grade essay questions or experiment reports, they have to read the answers in depth and
understand them. This should increase their level of understanding and knowledge, and reflect on their
answers in the final exams. The assignment will be phased in gradually into the courses (i.e., initially
they will not carry a lot of weight).
To evaluate the initial effects of this method, the grading assignment will be optional and will be
given after the midterm exams. Improvement in grades between midterms and the final exams for
students participating in the assignment will be compared to that of the other students
10
.
9 That is, to adjust the graded score according to the score of these questions that measure the level of understanding
of the material at hand.
10
Normal correlation between midterm exam grades and final exam grades (i.e., without this assignment) should also
be factored in.
Ahmed Helmy – USC – July 2001 (All rights reserved) 16 Student-assisted course development
To stimulate creativity in creating lab experiments, the students are asked to design new
experiments for current and future classes. Each student shall design one experiment individually and
two in a team. The students are also asked to provide sets of questions out of which a subset is chosen
by the instructor to be included in upcoming assignments or exams. Guidelines are given to students
to help develop high quality and relevant questions. These questions and their answers are graded for
quality.
On the first day of class a diagnostic exam is given that covers background and fundamentals as
well as elementary material to be taught in class. This gives indication of students’ weaknesses and
strengths. In addition, feedback forms are collected twice a semester (aside from conventional
evaluation at end of semester). Furthermore, short 5-minute exams, automatically computer-graded,
are given at the beginning and end of each lecture (called, in-class / out-of-class exams). Also, a novel
instant-feedback system will be used during each lecture. In such system, several (e.g., 4) multiple-
choice-questions are given by the instructor. Each student in class has a screen to view the questions
and answers, as well as a small device to enter the answer (a, b, c or d). The order of the questions and
answers on each screen is randomized to minimize cheating. The answers are automatically graded
11
. Statistics of these answers may be automatically correlated with overall student performance (e.g.,
GPA, diagnostic and midterm exam scores) to help identify troubled students and provide better
academic advice.
These procedures will be designed to measure the knowledge gained in class, to give students
motivation to pay attention in class and to raise the students’ level of understanding and education.
Broader Impact
The potential impacts of this research are significant. If successful, this research will provide the first
architecture for multicast service support in large-scale ad hoc networks, enabling a very important class
of applications in future networks. Our mechanisms are highly adaptive to network dynamics and
mobility , which renders our architecture more robust, efficient and scalable . Self-configurability of our
architecture facilitates ease-of-deployment. We also provide the first architecture for adaptive anycast in
ad hoc networks. Such mechanism can potentially provide the resource discovery component for a wide-
array of future applications and middle-ware. In addition, our systematic evaluation framework will
provide in-depth understanding of the effects of mobility on ad hoc network protocols (an effect that is
significant but not well-understood). One output of this research is a set of test-suites that can, potentially,
provide a meaningful and richer benchmarks for evaluating ad hoc networks, based on semantically rich
mobility models . Developed tools, simulation, implementation code, lab experiments, test scenarios, and
publications will be publicly available through the project’s web site. It will also be integrated into the
networking curriculum and courses at USC at both the undergraduate and graduate levels.
The research plan is complemented by a strong educational plan , potentially leading to significant
improvements in the learning process. The plan includes a wide-array of essential courses on Internet and
ad hoc networking protocols, and the establishment of a joint research/instruction wireless networking
laboratory . In addition, the PI provides a unique and effective approach to increase student involvement in
classroom activities and new methods for better evaluation of student abilities , using student-assisted
grading and course development. These methods will provide a model for improving education . References: [1] A. Helmy, "A Multicast-based Protocol for IP Mobility Support", ACM Second International Workshop on Networked Group Communication (NGC 2000), ACM SIGCOMM-sponsored, Palo Alto, November 2000.
[2] D. Estrin, D. Farinacci, A. Helmy, D. Thaler, S. Deering, V. Jacobson, M. Handley, C. Liu, P. Sharma, “Protocol Independent Multicast – Sparse Mode (PIM-SM): Protocol Specification”, RFC 2362/2117 of the Internet Engineering Task Force (IETF), Inter-Domain Multicast Routing (IDMR) , March '97/'98.
[3] S. Kumar, P. Radoslavov, D. Thaler, C. Alaettinoglu, D. Estrin, M. Handley, “The MASC/BGMP Architecture for Inter-domain Multicast Routing”, Proceedings of ACM SIGCOMM, August 1998.
[4] L. Breslau, D. Estrin, K. Fall, S. Floyd, J. Heidemann, A. Helmy, P. Huang, S. McCanne, K. Varadhan, Y. Xu, H. Yu, "Advances in Network Simulation", IEEE Computer, vol. 33, No. 5, p. 59-67, May 2000.
[5] A. Helmy, STRESS Systematic Test Synthesis for Multipoint Protocol Design, Ph.D. Dissertation, USC-CS-TR-99-716, Oct. 1999. [6] A. Helmy, D. Estrin, S. Gupta, Fault-oriented Test Generation for Multicast Routing Protocol Design, Proc. FORTE/PSTV ’98, IFIP, Nov. 1998.
[7] M. Zonoozi, P. Dassanayake, “User mobility modeling and characterization of mobility patterns”, IEEE Journal on Selected Areas in Communications, 1239-1252, September 1997.
[8] A. Bar-Noy, I. Kessler, M. Sidi, “Mobile users: to update or not to update?”, INFOCOM’94, pp 570-576, 1994.
[9] I. Rubin, C. Choi, “Impact of the location area structure on the performance of signaling channels in wireless cellular networks”, IEEE Communications Magazine, pp 108-115, February 1997. [10] H. Kobayashi, S. Yu, B. Mark, “An Integrated Mobility and Traffic Model for Resource Allocation in Wireless Networks”, WOWMOM, Boston, MA, 2000.
[11] M. Chung, T. Kim, H. Cho, D. Sung, “Modeling of Terminal to Evaluate the Number of Location Updates”, ICC’97, Montreal, 1997.
[12] X. Hong, M. Gerla, G. Pei, C. Chiang, “A group mobility model for ad hoc wireless networks”, 2 nd ACM Int’l workshop on modeling, analysis and simulation of wireless and mobile systems, pp 53-60, 1999
[13] A. McDonald, T. Znati, “A mobility-based framework for adaptive clustering in wireless ad hoc networks”, IEEE Journal on Selected Areas in Communications, V. 17 (8), pp 1466-1487, Aug. 1999.
[14] D. Estrin, M. Handley, A. Helmy, P. Huang, D. Thaler, "A Dynamic Bootstrap Mechanism for Rendezvous-based Multicast Routing", Proceedings of IEEE INFOCOM '99, New York, March 1999. [15] S. Deering, D. Cheriton, “Multicast routing in data internetworks and extended lans”, ACM Transactions on Computer Systems, pp 85-111, May 1990.
[16] A. Helmy, S. Gupta, D. Estrin, A. Cerpa, Y. Yu, "Systematic Performance Evaluation of Multipoint Protocols", Proceedings of Formal Description Techniques & Protocol Specification, Testing, and Verification (FORTE/PSTV), IFIP, Kluwer Academic Publication, Pisa, Italy, p. 187-202, October 2000.
[17] Kevin Fall. Network Emulation in the Vint/NS Simulator, ISCC '99.
[18] C. E. Perkins, P. Bhagwat, Highly Dynamic Destination-Sequenced Distance Vector Routing (DSDV) for Mobile Computers, Comp. Commun. Rev., Oct. 1994, pp. 234-44.
[19] C.-C. Chiang, Routing in Clustered Multihop, Mobile Wireless Networks with Fading Channel, Proc. IEEE SICON ’97, pp. 197-211.
[20] S. Murthy, J. J. Garcia-Luna-Aceves, An Efficient Routing Protocol for Wireless Networks, ACM Mobile Networks and App. J., Special Issue on Routing in Mobile Communication Networks, Oct. 1996, pp. 183-197.
[21] C. E. Perkins, E. M. Royer, Ad-hoc On-Demand Distance Vector Routing, Proc. 2 nd IEEE Wksp. Mobile Comp. Sys. And Apps., Feb. 1999, pp. 90-100.
[22] D. B. Johnson, D. A. Maltz, Dynamic Source Routing in Ad-Hoc Wireless Networks, Mobile Computing, 1996, pp.153-181.
[23] V.D. Park, M.S. Corson, A Highly Adaptive Distributed Routing Algorithm for Mobile Wireless Networks, Proc. INFOCOM ’97, Apr. 1997.
[24] C-K. Toh, A Novel Distributed Routing Protocol To Support Ad-Hoc Mobile Computing, Proc. 1996 IEEE 15 th Annual Int’l. Phoenix Conf. Comp. And Commun., Mr. 1996, pp. 480-486.
[25] R. Dube et al., Signal Stability based Adaptive Routing (SSA) for Ad-Hoc Mobile Networks, IEEE Pers. Commun., Feb. 1997, pp. 36-45.
[26] J. J. Garcia-Luna-Aceves, E. L. Madruga, A Multicast Routing Protocol for Ad-Hoc Networks, Proc. IEEE INFOCOM ‘99, pp. 784-792, 1999.
[27] M. Gerla, C.-C. Chiang, L. Zhang, Tree Multicast Strategies in Mobile, Multihop Wireless Networks, ACM Mobile Networks and Apps. J., 1998.
[28] Z. Haas, “A new routing protocol for the reconfigurable wireless networks”, IEEE Conference on Universal Personal Communications, Vol. 2, pp. 562-566, 1997.
[29] D. J. Watts. In Small Worlds, The dynamics of networks between order and randomnes s. Princeton University Press, 1999.
[30] L. Blazevic, S. Giordano, J. Boudec, “self-organizing wide-area routing”, SCI/ISAS July 2000,
[31] D. Katabi, J. Wroclawski, “A framework for scalable global IP anycast (GIA), Proceedings of ACM Sigcomm 2000.
[32] P. F. Tsuchiya, “The Landmark Hierarchy: A new hierarchy for routing in very large networks”, Computer Communication Review, Vol. 18, no. 4, pp. 35-42, Aug. 1988.
[33] E. Bommaiah, M. Liu, A. McAuley, R. Talpade, “Ad-hoc Multicast Routing Protocol”, Internet-draft, Aug. 1998.
[34] C. Wu, Y. Tay, C. Toh, “Ad hoc Multicast Routing Protocol utilizing Increasing id-numbers (AMRIS): functional specification”, Internet-draft, November 1998.
[35] Y. Ko, N. Vaidya, “Location-aided routing (LAR) in mobile ad hoc networks”, Wireless Networks 6, 4, p. 307-321, July 2000.
[36] S. Lee, M. Gerla, C. Chiang, “On-demand multicast routing protocol”, IEEE WCNC, p. 1298-1302, vol. 3, 1999.
[37] S. Lee, W. Su, M. Gerla, “Ad hoc wireless multicast with mobility prediction”, IEEE ICCN, p. 4-9, 1999.
[38] P. Guangyu, M. Gerla, X. Hong, “LANMAR: landmark routing for large scale wireless ad hoc networks with group mobility”, MobiHOC ’00, p. 11-18, 2000.
[39] C. Chiang, M. Gerla, L. Zhang, “Forwarding Group Multicast Protocol (FGMP) for Multihop, Mobile Wireless Networks”, ACM/Kluwer Journal of Cluter Computing, vol. 1, no. 2, 1998.
[40] R. Sivakumar, P. Sinha, V. Bharghavan, “CEDAR: a core-extraction distributed ad hoc routing algorithm”, IEEE JSAC vol. 17, 8, p. 1454-1465, Aug. 1999.
[41] P. Sinha, R. Sivakumar, V. Bharghavan, “MCEDAR: multicast core-extraction distributed ad hoc routing”, IEEE WCNC, p. 1313-1317, 1999.
[42] D. Estrin, D. Farinacci, A. Helmy, V. Jacobson, L. Wei, “Protocol Independent Multicast – Dense Mode (PIM-DM): Protocol Specification”, Proposed RFC of the IETF/IDMR , September 1996.
[43] Y. Ko, N. Vaidya, “Geocasting in mobile ad hoc networks: location-based multicast algorithms”, IEEE WMCSA, p. 101-110, 1999.
[44] Y. Ko, N. Vaidya, “GeoTORA: a protocol for geocasting in mobile ad hoc networks”, IEEE ICNP, p. 240-250, 2000.
[45] S. Deering, W. Fenner, D. Estrin, A. Helmy, D. Farinacci, L. Wei, M. Handley, V. Jacobson, D. Thaler, “Hierarchical PIM for Inter-Domain Multicast Routing”, Internet-Draft (I-D) of the IETF/IDMR , December 1995.
[46] A. Helmy, D. Estrin, S. Gupta, “Systematic Testing of Multicast Routing Protocols: Analysis of Forward and Backward Search Techniques”, IEEE ICCCN, p. 590-597, October 2000.
[47] J. J. Aceves, E. Madruga, “The Core-Assisted Mesh Protocol”, IEEE Journal on Selected Areas in Communications, vol. 17, no. 8, pp. 1380-1394, August 1999.
[48] A. Nasipuri, S. Das, “On-Demand Multipath Routing for Mobile Ad Hoc Networks”, ICCN, p. 64-70, 1999.
[49] S. Das, A. Mukherjee, S. Bandyopadhyay, K. Paul, D. Saha, “Improving quality-of-service in ad hoc wireless networks with adaptive multi-path routing”, GlobeCom, p. 261-265, vol. 1, 2000.
[50] S. Kumar, C. Alaettinoglu, D. Estrin, “Scalable Object-tracking through Unattended Techniques (SCOUT)”, IEEE ICNP, Nov. 2000.
[51] M. Ng, I Lu, “A peer-to-peer zone-based two-level link state routing for mobile ad hoc networks”, IEEE Journal on Selected Areas in Communications, Vol. 17, no. 8, pp. 1415-1425, Aug. 1999.
[52] M. Pearlman, Z. Haas, “Determining the optimal configuration for the zone routing protocol”, IEEE JSAC, p. 1395-1414, vol. 17, 8, Aug 1999.
[53] A. McDonald, T. Znati, “A path availability model for wireless ad-hoc networks”, IEEE WCNC, p. 35-40, vol. 1, 1999.
[54] A. McDonald, T. Znati, “Predicting node proximity in ad-hoc networks: a least overhead adaptive model for selecting stable routes”, MobiHOC, p. 29-33, 2000.
[55] D. Watts, S. Strogatz, "Collective dynamics of 'small-world' networks", Nature, Vol. 393, June 4, 1998.
[56] M. Ize, Y. Kim, “PatchODMRP: an ad-hoc multicast routing protocol”, ICIN, p. 537-543, 2001.
[57] P. Basu, N. Khan, T. Little, “A mobility based metric for clustering in mobile ad hoc networks”, ICDCS Workshop, p. 413-418, 2001.
[58] N. Bulusu, J. Heidemann, D. Estrin, “GPS-less Low Cost Outdoor Localization For Very Small Devices”, IEEE Personal Communications, Vol. 7, 5, p. 28-34, October 2000.
[59] S. Capkun, M. Hamdi, J. Hubaux, “GPS-free positioning in mobile ad-hoc networks”, ICSC, p. 3481-3490, 2001.
[60] D. Estrin, D. Farinacci, A. Helmy, D. Thaler, S. Deering, V. Jacobson, M. Handley, C. Liu, P. Sharma, L. Wei, "Protocol Independent Multicast (PIM): Motivation and Architecture", Proposed RFC of the IETF/IDMR, October 1996.
[61] A. Battacharya, S. Das, “LeZi-Update: An Information-Theoretic Approach to Track Mobile Users in PCS Networks”, MobiCom ’99.
[62] A. Helmy, " State Analysis and Aggregation Study for Multicast-based Micro Mobility", ACM-LANL-NCSTRL cs.NI/0105031, May 2001.
[63] H. Adeli, A. Karim, “Fuzzy-Wavelet RBFNN Model for Freeway Incident Detection”, Journal of Transportation Engineering, Dec. 2000.
[64] H. Zhang, “Recursive Prediction of Traffic Conditions with Neural Networks Models”, Journal of Transportation Engineering, Dec. 2000.
[65] G. Pei, M. Gerla, T. Chen, “Fisheye State Routing: A Routing Scheme for Ad Hoc Wireless Networks”, ICC 2000.

11
A similar concept will also be used to provide instant feedback on instructor performance (e.g., on volume of
voice, clarity or repetition, etc.).
Ahmed Helmy – USC – July 2001 (All rights reserved) 17 References:
[1] A. Helmy, "A Multicast-based Protocol for IP Mobility Support", ACM Second International
Workshop on Networked Group Communication (NGC 2000), ACM SIGCOMM-sponsored, Palo
Alto, November 2000.
[2] D. Estrin, D. Farinacci, A. Helmy, D. Thaler, S. Deering, V. Jacobson, M. Handley, C. Liu, P.
Sharma, “Protocol Independent Multicast – Sparse Mode (PIM-SM): Protocol Specification”, RFC
2362/2117 of the Internet Engineering Task Force (IETF), Inter-Domain Multicast Routing (IDMR) , March '97/'98.
[3] S. Kumar, P. Radoslavov, D. Thaler, C. Alaettinoglu, D. Estrin, M. Handley, “The MASC/BGMP
Architecture for Inter-domain Multicast Routing”, Proceedings of ACM SIGCOMM, August 1998.
[4] L. Breslau, D. Estrin, K. Fall, S. Floyd, J. Heidemann, A. Helmy, P. Huang, S. McCanne, K.
Varadhan, Y. Xu, H. Yu, "Advances in Network Simulation", IEEE Computer, vol. 33, No. 5, p. 59-
67, May 2000.
[5] A. Helmy, STRESS Systematic Test Synthesis for Multipoint Protocol Design, Ph.D. Dissertation,
USC-CS-TR-99-716, Oct. 1999.
[6] A. Helmy, D. Estrin, S. Gupta, Fault-oriented Test Generation for Multicast Routing Protocol Design,
Proc. FORTE/PSTV ’98, IFIP, Nov. 1998.
[7] M. Zonoozi, P. Dassanayake, “User mobility modeling and characterization of mobility patterns”,
IEEE Journal on Selected Areas in Communications, 1239-1252, September 1997.
[8] A. Bar-Noy, I. Kessler, M. Sidi, “Mobile users: to update or not to update?”, INFOCOM’94, pp 570-
576, 1994.
[9] I. Rubin, C. Choi, “Impact of the location area structure on the performance of signaling channels in
wireless cellular networks”, IEEE Communications Magazine, pp 108-115, February 1997.
[10] H. Kobayashi, S. Yu, B. Mark, “An Integrated Mobility and Traffic Model for Resource
Allocation in Wireless Networks”, WOWMOM, Boston, MA, 2000.
[11] M. Chung, T. Kim, H. Cho, D. Sung, “Modeling of Terminal to Evaluate the Number of Location
Updates”, ICC’97, Montreal, 1997.
[12] X. Hong, M. Gerla, G. Pei, C. Chiang, “A group mobility model for ad hoc wireless networks”,
2 nd
ACM Int’l workshop on modeling, analysis and simulation of wireless and mobile systems, pp 53-
60, 1999
[13] A. McDonald, T. Znati, “A mobility-based framework for adaptive clustering in wireless ad hoc
networks”, IEEE Journal on Selected Areas in Communications, V. 17 (8), pp 1466-1487, Aug. 1999.
[14] D. Estrin, M. Handley, A. Helmy, P. Huang, D. Thaler, "A Dynamic Bootstrap Mechanism for
Rendezvous-based Multicast Routing", Proceedings of IEEE INFOCOM '99, New York, March 1999.
[15] S. Deering, D. Cheriton, “Multicast routing in data internetworks and extended lans”, ACM
Transactions on Computer Systems, pp 85-111, May 1990.
[16] A. Helmy, S. Gupta, D. Estrin, A. Cerpa, Y. Yu, "Systematic Performance Evaluation of
Multipoint Protocols", Proceedings of Formal Description Techniques & Protocol Specification,
Testing, and Verification (FORTE/PSTV), IFIP, Kluwer Academic Publication, Pisa, Italy, p. 187-
202, October 2000.
[17] Kevin Fall. Network Emulation in the Vint/NS Simulator, ISCC '99.
[18] C. E. Perkins, P. Bhagwat, Highly Dynamic Destination-Sequenced Distance Vector Routing
(DSDV) for Mobile Computers, Comp. Commun. Rev., Oct. 1994, pp. 234-44.
[19] C.-C. Chiang, Routing in Clustered Multihop, Mobile Wireless Networks with Fading Channel,
Proc. IEEE SICON ’97, pp. 197-211.
[20] S. Murthy, J. J. Garcia-Luna-Aceves, An Efficient Routing Protocol for Wireless Networks, ACM
Mobile Networks and App. J., Special Issue on Routing in Mobile Communication Networks, Oct.
1996, pp. 183-197.
[21] C. E. Perkins, E. M. Royer, Ad-hoc On-Demand Distance Vector Routing, Proc. 2
nd
IEEE Wksp.
Mobile Comp. Sys. And Apps., Feb. 1999, pp. 90-100.
Ahmed Helmy – USC – July 2001 (All rights reserved) 18 [22] D. B. Johnson, D. A. Maltz, Dynamic Source Routing in Ad-Hoc Wireless Networks, Mobile
Computing, 1996, pp.153-181.
[23] V.D. Park, M.S. Corson, A Highly Adaptive Distributed Routing Algorithm for Mobile Wireless
Networks, Proc. INFOCOM ’97, Apr. 1997.
[24] C-K. Toh, A Novel Distributed Routing Protocol To Support Ad-Hoc Mobile Computing, Proc.
1996 IEEE 15
th
Annual Int’l. Phoenix Conf. Comp. And Commun., Mr. 1996, pp. 480-486.
[25] R. Dube et al., Signal Stability based Adaptive Routing (SSA) for Ad-Hoc Mobile Networks,
IEEE Pers. Commun., Feb. 1997, pp. 36-45.
[26] J. J. Garcia-Luna-Aceves, E. L. Madruga, A Multicast Routing Protocol for Ad-Hoc Networks,
Proc. IEEE INFOCOM ‘99, pp. 784-792, 1999.
[27] M. Gerla, C.-C. Chiang, L. Zhang, Tree Multicast Strategies in Mobile, Multihop Wireless
Networks, ACM Mobile Networks and Apps. J., 1998.
[28] Z. Haas, “A new routing protocol for the reconfigurable wireless networks”, IEEE Conference on
Universal Personal Communications, Vol. 2, pp. 562-566, 1997.
[29] D. J. Watts. In Small Worlds, The dynamics of networks between order and randomnes s.
Princeton University Press, 1999.
[30] L. Blazevic, S. Giordano, J. Boudec, “self-organizing wide-area routing”, SCI/ISAS July 2000,
[31] D. Katabi, J. Wroclawski, “A framework for scalable global IP anycast (GIA), Proceedings of
ACM Sigcomm 2000.
[32] P. F. Tsuchiya, “The Landmark Hierarchy: A new hierarchy for routing in very large networks”,
Computer Communication Review, Vol. 18, no. 4, pp. 35-42, Aug. 1988.
[33] E. Bommaiah, M. Liu, A. McAuley, R. Talpade, “Ad-hoc Multicast Routing Protocol”, Internet-
draft, Aug. 1998.
[34] C. Wu, Y. Tay, C. Toh, “Ad hoc Multicast Routing Protocol utilizing Increasing id-numbers
(AMRIS): functional specification”, Internet-draft, November 1998.
[35] Y. Ko, N. Vaidya, “Location-aided routing (LAR) in mobile ad hoc networks”, Wireless
Networks 6, 4, p. 307-321, July 2000.
[36] S. Lee, M. Gerla, C. Chiang, “On-demand multicast routing protocol”, IEEE WCNC, p. 1298-
1302, vol. 3, 1999.
[37] S. Lee, W. Su, M. Gerla, “Ad hoc wireless multicast with mobility prediction”, IEEE ICCN, p. 4-
9, 1999.
[38] P. Guangyu, M. Gerla, X. Hong, “LANMAR: landmark routing for large scale wireless ad hoc
networks with group mobility”, MobiHOC ’00, p. 11-18, 2000.
[39] C. Chiang, M. Gerla, L. Zhang, “Forwarding Group Multicast Protocol (FGMP) for Multihop,
Mobile Wireless Networks”, ACM/Kluwer Journal of Cluter Computing, vol. 1, no. 2, 1998.
[40] R. Sivakumar, P. Sinha, V. Bharghavan, “CEDAR: a core-extraction distributed ad hoc routing
algorithm”, IEEE JSAC vol. 17, 8, p. 1454-1465, Aug. 1999.
[41] P. Sinha, R. Sivakumar, V. Bharghavan, “MCEDAR: multicast core-extraction distributed ad hoc
routing”, IEEE WCNC, p. 1313-1317, 1999.
[42] D. Estrin, D. Farinacci, A. Helmy, V. Jacobson, L. Wei, “Protocol Independent Multicast – Dense
Mode (PIM-DM): Protocol Specification”, Proposed RFC of the IETF/IDMR , September 1996.
[43] Y. Ko, N. Vaidya, “Geocasting in mobile ad hoc networks: location-based multicast algorithms”,
IEEE WMCSA, p. 101-110, 1999.
[44] Y. Ko, N. Vaidya, “GeoTORA: a protocol for geocasting in mobile ad hoc networks”, IEEE
ICNP, p. 240-250, 2000.
[45] S. Deering, W. Fenner, D. Estrin, A. Helmy, D. Farinacci, L. Wei, M. Handley, V. Jacobson, D.
Thaler, “Hierarchical PIM for Inter-Domain Multicast Routing”, Internet-Draft (I-D) of the
IETF/IDMR , December 1995.
[46] A. Helmy, D. Estrin, S. Gupta, “Systematic Testing of Multicast Routing Protocols: Analysis of
Forward and Backward Search Techniques”, IEEE ICCCN, p. 590-597, October 2000.
Ahmed Helmy – USC – July 2001 (All rights reserved) 19 [47] J. J. Aceves, E. Madruga, “The Core-Assisted Mesh Protocol”, IEEE Journal on Selected Areas in
Communications, vol. 17, no. 8, pp. 1380-1394, August 1999.
[48] A. Nasipuri, S. Das, “On-Demand Multipath Routing for Mobile Ad Hoc Networks”, ICCN, p. 64-70, 1999.
[49] S. Das, A. Mukherjee, S. Bandyopadhyay, K. Paul, D. Saha, “Improving quality-of-service in ad hoc
wireless networks with adaptive multi-path routing”, GlobeCom, p. 261-265, vol. 1, 2000.
[50] S. Kumar, C. Alaettinoglu, D. Estrin, “Scalable Object-tracking through Unattended Techniques
(SCOUT)”, IEEE ICNP, Nov. 2000.
[51] M. Ng, I Lu, “A peer-to-peer zone-based two-level link state routing for mobile ad hoc networks”, IEEE
Journal on Selected Areas in Communications, Vol. 17, no. 8, pp. 1415-1425, Aug. 1999.
[52] M. Pearlman, Z. Haas, “Determining the optimal configuration for the zone routing protocol”, IEEE JSAC,
p. 1395-1414, vol. 17, 8, Aug 1999.
[53] A. McDonald, T. Znati, “A path availability model for wireless ad-hoc networks”, IEEE WCNC, p. 35-40,
vol. 1, 1999.
[54] A. McDonald, T. Znati, “Predicting node proximity in ad-hoc networks: a least overhead adaptive model for
selecting stable routes”, MobiHOC, p. 29-33, 2000.
[55] D. Watts, S. Strogatz, "Collective dynamics of 'small-world' networks", Nature, Vol. 393, June 4, 1998.
[56] M. Ize, Y. Kim, “PatchODMRP: an ad-hoc multicast routing protocol”, ICIN, p. 537-543, 2001.
[57] P. Basu, N. Khan, T. Little, “A mobility based metric for clustering in mobile ad hoc networks”,
ICDCS Workshop, p. 413-418, 2001.
[58] N. Bulusu, J. Heidemann, D. Estrin, “GPS-less Low Cost Outdoor Localization For Very Small
Devices”, IEEE Personal Communications, Vol. 7, 5, p. 28-34, October 2000.
[59] S. Capkun, M. Hamdi, J. Hubaux, “GPS-free positioning in mobile ad-hoc networks”, ICSC, p.
3481-3490, 2001.
[60] D. Estrin, D. Farinacci, A. Helmy, D. Thaler, S. Deering, V. Jacobson, M. Handley, C. Liu, P.
Sharma, L. Wei, "Protocol Independent Multicast (PIM): Motivation and Architecture", Proposed
RFC of the IETF/IDMR, October 1996.
[61] A. Battacharya, S. Das, “LeZi-Update: An Information-Theoretic Approach to Track Mobile
Users in PCS Networks”, MobiCom ’99.
[62] A. Helmy, " State Analysis and Aggregation Study for Multicast-based Micro Mobility", ACM-
LANL-NCSTRL cs.NI/0105031, May 2001.
[63] H. Adeli, A. Karim, “Fuzzy-Wavelet RBFNN Model for Freeway Incident Detection”, Journal of
Transportation Engineering, Dec. 2000.
[64] H. Zhang, “Recursive Prediction of Traffic Conditions with Neural Networks Models”, Journal of
Transportation Engineering, Dec. 2000.
[65] G. Pei, M. Gerla, T. Chen, “Fisheye State Routing: A Routing Scheme for Ad Hoc Wireless
Networks”, ICC 2000.
[66]

Linked assets

Computer Science Technical Report Archive

Conceptually similar

PDF

USC Computer Science Technical Reports, no. 749 (2001)

PDF

USC Computer Science Technical Reports, no. 780 (2002)

PDF

USC Computer Science Technical Reports, no. 804 (2003)

PDF

USC Computer Science Technical Reports, no. 770 (2002)

PDF

USC Computer Science Technical Reports, no. 781 (2002)

PDF

USC Computer Science Technical Reports, no. 790 (2003)

PDF

USC Computer Science Technical Reports, no. 860 (2005)

PDF

USC Computer Science Technical Reports, no. 801 (2003)

PDF

USC Computer Science Technical Reports, no. 734 (2000)

PDF

USC Computer Science Technical Reports, no. 789 (2003)

PDF

USC Computer Science Technical Reports, no. 755 (2002)

PDF

USC Computer Science Technical Reports, no. 816 (2004)

PDF

USC Computer Science Technical Reports, no. 753 (2002)

PDF

USC Computer Science Technical Reports, no. 747 (2001)

PDF

USC Computer Science Technical Reports, no. 809 (2003)

PDF

USC Computer Science Technical Reports, no. 757 (2002)

PDF

USC Computer Science Technical Reports, no. 765 (2002)

PDF

USC Computer Science Technical Reports, no. 775 (2002)

PDF

USC Computer Science Technical Reports, no. 772 (2002)

PDF

USC Computer Science Technical Reports, no. 788 (2003)

Description

Ahmed Helmy. "Adaptive architectures for multicast service support in large-scale mobile ad hoc networks: Design and evaluation framework." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 803 (2003).

Asset Metadata

Creator Helmy, Ahmed (author)

Core Title USC Computer Science Technical Reports, no. 803 (2003)

Alternative Title Adaptive architectures for multicast service support in large-scale mobile ad hoc networks: Design and evaluation framework (title)

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)

Tag OAI-PMH Harvest

Format 19 pages (extent), technical reports (aat)

Language English

Unique identifier UC16269675

Identifier 03-803 Adaptive Architectures for Multicast Service Support in Large-Scale Mobile Ad Hoc Networks Design and Evaluation Framework (filename)

Legacy Identifier usc-cstr-03-803

Format 19 pages (extent),technical reports (aat)

Rights Department of Computer Science (University of Southern California) and the author(s).

Internet Media Type application/pdf

Source 20180426-rozan-cstechreports-shoaf (batch), Computer Science Technical Report Archive (collection), University of Southern California. Department of Computer Science. Technical Reports (series)

Access Conditions The author(s) retain rights to their work according to U.S. copyright law. Electronic access is being provided by the USC Libraries, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Repository Email csdept@usc.edu

Inherited Values

Title Computer Science Technical Report Archive

Description Archive of computer science technical reports published by the USC Department of Computer Science from 1991 - 2017.

Coverage Temporal 1991/2017

Repository Email csdept@usc.edu

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)