Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
A Robust Routing Strategy for Density Spanner based Wireless Ad Hoc
Networks
Divi Mydhili1, K.R.R.Mohana Rao2
1Pursuing M.Tech(CS), Nalanda Institute of Engineering & Technology,Siddharth Nagar, Sattenapalli,
Guntur., Affiliated to JNTUK, Kakinada, A.P., India.
2Professor & Head, Department of Computer Science Engineering, Nalanda Institute of Engineering &
Technology, Siddharth Nagar, Sattenapalli, Guntur., Affiliated to JNTUK, Kakinada, A.P., India.
Abstract - An important problem for wireless ad hoc
networks has been to design overlay networks that
allow time- and energy-efficient routing. Many local-
control strategies for maintaining such overlay
networks have already been suggested, but most of
them are based on an oversimplified wireless
communication model. We address cooperative caching
in wireless networks, where the nodes may be mobile
and exchange information in a peer-to-peer fashion.
We consider both cases of nodes with large and small-
sized caches. For large-sized caches, we devise a
strategy where nodes, independent of each other,
decide whether to cache some content and for how
long. In the case of small-sized caches, we aim to
design a content replacement strategy that allows
nodes to successfully store newly received information
while maintaining the good performance of the content
distribution system. Under both conditions, each node
takes decisions according to its perception of what
nearby users may store in their caches and with the aim
of differentiating its own cache content from the other
nodes’. In this paper, we suggest a model that is much
more general than previous models. It allows the path
loss of transmissions to significantly deviate from the
idealistic unit disk model and does not even require the
path loss to form a metric. Also, our model is
apparently the first proposed for algorithm designs that
does not only model transmission and interference
issues but also aims at providing a realistic model for
physical carrier sensing. Physical carrier sensing is
needed so that our protocols do not require any prior
information (not even an estimate on the number of
nodes) about the wireless network to run efficiently.
Keywords - Routing, Protocols, ad hoc networks,
spanner, dominating set.
1.Introduction Providing information to users on the move is one of
the most promising directions of the infotainment
business, which rapidly becomes a market reality,
because infotainment modules are deployed on cars and
handheld devices. The ubiquity and ease of access of
third- and fourth-generation (3G or 4G) networks will
encourage users to constantly look for content that
matches their interests. However, by exclusively
relying on downloading from the infrastructure, novel
applications such as mobile multimedia are likely to
overload the wireless network (as recently happened to
AT&T following the introduction of the iPhone [1]). It
is thus conceivable that a peer-to-peer system could
come in handy, if used in conjunction with cellular
networks, to promote content sharing using ad hoc
networking among mobile users [2]. Any cast is an
addressing mode in which the same address is assigned
to multiple hosts. Together, these hosts form an any
cast group and each host is referred to as an any cast
group member.
Packets from a client destined to the group
address are routed to the any cast group member closest
to the client, where ”closest” is in terms of the metrics
used by the underlying routing protocol. The most
prominent use of any cast today is in the Internet to find
replicated DNS root servers or to locate rendezvous
points in multicast trees [3]. However, any cast has also
many potential applications in wireless ad hoc
networks. For example, any cast can be used in wireless
mesh networks to route data packets to an Internet
gateway, or in sensor networks to send data to any data
sink when multiple sinks are accessible. Today’s any
cast routing protocols are most commonly
modifications of existing unicast routing protocols. For
example, link-state routing protocols such as OSPF [4]
have been extended to support any cast routing by
adding a virtual node that represents the any cast
service [5]. With distance vector routing algorithms
such as RIP [6], any cast routing is implemented by
group members that advertise their any cast address
with a distance of zero [5]. Also in the context of ad
hoc networks, link reversal algorithms such as TORA
[7] were extended to support any cast routing by
assigning a height of zero to all members of a given any
cast group [5]. Since these proposed any cast protocols
are designed as extensions of unicast routing
techniques, they are easy to implement and to deploy.
Divi Mydhili et al ,Int.J.Computer Technology & Applications,Vol 3 (1), 396-400
IJCTA | JAN-FEB 2012 Available [email protected]
396
ISSN:2229-6093
However, as a consequence, they all follow the routing
strategy determined by the corresponding unicast
routing technique: packet delivery to the closest group
member using shortest-path forwarding.
2. Anycast Routing Model
In this section, we present our field-based model for
any cast routing. First, we give an overview of the basic
concepts. Then, we introduce our definition of potential
fields in a networking context and describe how
packets are routed along those fields. Finally, we
discuss convergence limitations of the model due to
undesired local maxima in the fields that might occur in
particular network topologies.
Overview:
Our model is inspired from field theory in physics.
Every group member creates a potential field which
decreases with d−k
, where d is the distance to the group
member, and k determines how quickly the field
decreases. The field of an entire anycast group is the
linear superposition of all individual fields from the
group members. An example field for an anycast group
with four members (marked as black nodes) is depicted
in Figure 1. The peaks in the field are at the locations of
the group members. Note that only one field is drawn in
the figure, but as each anycast group requires its own
field, multiple fields will generally co-exist
simultaneously. We achieve anycast routing by
forwarding packets towards the steepest gradient of the
field. This is in analogy to field diffusion in physics. By
following the steepest gradient, packets eventually
reach a field maximum, i.e., a group member. The
steepest gradient at each node is determined by
comparing the potential values ϕ of its neighbors. The
steepest gradient is towards the neighbor with the
highest potential value.
Fig. 1. Example potential field. Black nodes represent
group members.
The proposed routing model allows for different
anycast strategies comprising proximity, density, and
combined routing strategies. Which routing strategy is
applied is determined by the value of the parameter k.
We will show in the next section that a proximity-based
routing strategy (the routing strategy of existing anycast
routing protocols which consists of forwarding packets
to the closest member along the shortest path) is
modeled by setting k > μ, where μ is a constant
depending on the network size and the anycast group
size. We also show that for 0 < k ≤ ǫ (where ǫ <
μ), a pure density based routing strategy is modelled
where the proximity of the group members is no longer
considered for routing decisions. By choosing a value
for k between μ and ǫ, we are able to model
combinations of these two one-sided routing strategies.
3. Field-Based Anycast Routing
Protocol To evaluate the performance of density-based anycast
routing in dynamic networks, we designed a distributed
routing protocol to establish potential fields and
forward packets along the steepest gradient. Note that
the focus of this paper is not on the performance of the
routing protocol itself, but on the comparison of the
different anycast routing strategies.
Therefore, we designed a relatively simple protocol for
the only purpose of comparing the different strategies,
and leave possible enhancements of the protocol to
future work.
A. Potential field establishment
To establish a potential field, every node in the network
must know its distance to the existing group members.
For this purpose, the group members periodically flood
the network with a message indicating the anycast
group they belong to, and their identity (i.e., an
identifier that uniquely identifies the group member).
These flooded messages also include a time to live
(TTL) value indicating how many hops the packet has
traveled. The TTL value serves two purposes. First, it
allows every receiver to determine its distance to the
group member who initiated the flooding. Second, it
allows to limit the flooding scope by only rebroad
casting messages which have a TTL value greater than
0 which reduces the communication overhead produced
by such messages. By listening to those messages, each
node calculates its potential value according to
Equation (2). The interval at which the member should
advertise such messages is a tradeoff between accuracy
and protocol overhead. In this paper, we do not focus
on finding the best compromise with regard to this
tradeoff, but we study the relative performance
resulting from different anycast strategies using the
same advertisement interval. Note that the impact of the
Divi Mydhili et al ,Int.J.Computer Technology & Applications,Vol 3 (1), 396-400
IJCTA | JAN-FEB 2012 Available [email protected]
397
ISSN:2229-6093
TTL value on the performance of the routing is
evaluated.
B. Gradient determination
To determine the steepest gradient of a field, the nodes
in an ad hoc network must know the potential values of
their direct neighbors. For this, neighboring nodes also
exchange their potential values on a periodic basis.
Such messages are one-hop broadcast packets and
include a list of all the known anycast groups and the
corresponding potential value for each group. Again,
the rate at which such messages are exchanged is a
tradeoff between accuracy and protocol overhead.
C. Packet forwarding
Packet forwarding is simply forwarding along the
steepest gradient. Therefore, packets are forwarded to
the neighbor with the highest potential value. If for any
reason, the neighbor with the greatest potential value is
unreachable (e.g., this neighbor might have moved
away), the packet is simply forwarded to the neighbor
with the second highest potential value. In case this
node is also unreachable, the packet is forwarded to the
neighbor with the next highest potential value, and so
on. A node continues with this procedure until there are
no neighbors left with a higher potential value than its
own. Note that nodes are not allowed to forward to a
neighbor with a lower potential value to make sure that
routing eventually converges and loops do not form. If
a node has no neighbors left with a larger potential
value than its own, it drops the packet.
4. Constant Density Spanner
In the next two subsections, we describe phases II and
III in detail. We use the following notation. The
constant d1 refers to the number of active nodes that
are within the interference range ri of any node. The
constant d2 refers to the number of active nodes that
are within the ri ri-range of any node, and the
constant g refers to the maximum number of required
gateway connections for any active node. Finally, D
refers to the density of the network, i.e. the maximum
number of nodes within the transmission range of a
node.
4.1 Phase II - Distributed Leader Coloring
Similar to phase I, each node organizes the time into
time frames consisting of cd1 rounds for some constant
c that is the same for every node. Also here, the rounds
are synchronized but frames do not have to be
synchronized among the nodes. We again assign active
nodes to distinct rounds using a coloring mechanism.
While the coloring in phase I was done with respect to
Grt , we now need a coloring of the active nodes with
respect to Gri ri , that is, we need the active nodes to
be at least ri ri apart in order to receive the same
color. Every active node from phase I tries to own one
of the rounds.
An active node u is said to own a round if no other
active node within its ri ri range is using that round.
Active nodes are in one of the states {owner, volatile}.
An active node is in owner state if it already owns a
round and is in volatile state if it is still trying to own a
round. Active nodes in owner state always send their
ID in the first time slot of their round. Initially, every
active node is volatile. Active nodes in volatile state
choose an active round from the cd1 possible rounds
uniformly at random. Active nodes in owner state use a
sensing threshold To with CST range ri and active
nodes in volatile state use a sensing threshold Tv with a
CST range being equal to the CSI range of To, rii.
Active nodes do the following repeatedly. Every time a
node reactivates, it sets its time stamp to 0. This time
stamp is used by active nodes in Phase III to compare
entries.
1. Every active node in owner state that is in its active
round sends out a LEADER message containing its ID
and its current time stamp and increases its time stamp
by one afterwards.
2. Every active node in owner state that is in its active
round decides with probability 1/2 to send out an
OWNER message either in the first or second substep
of step 2.
3. Every inactive node that sensed a LEADER message
with threshold Tv sends out a BUSY signal. Every
active node in volatile state that senses a BUSY signal
in its active round chooses a new active round
uniformly at random.
4. Every inactive node that sensed OWNER messages
in both substeps of step 2 with threshold To sends out a
COLLISION signal.
If an active node in owner state senses a COLLISION
signal and sent an OWNER message in the second
substep, it changes into volatile state and chooses a new
active round uniformly at random. If an active node in
volatile state did not sense a BUSY or COLLISION
signal in its active round, it becomes an owner.
THEOREM4.1. Once a stable set of active nodes is
available, it holds: If c ≥ 4, then all active nodes will be
in owner state after O(log n) rounds of the protocol,
w.h.p. The theorem implies that after O(log n) rounds,
Divi Mydhili et al ,Int.J.Computer Technology & Applications,Vol 3 (1), 396-400
IJCTA | JAN-FEB 2012 Available [email protected]
398
ISSN:2229-6093
all active nodes have chosen rounds so that for any two
active nodes L and L` with the same round and any
inactive node v within the interference range of L,L` is
outside of the interference range of v. Hence, L can
transmit messages to nodes within its transmission
range without interference problems, and these nodes
can transmit messages to L without causing
interference problems at L. Both properties are
important for phase III to work correctly.
4.2 Phase III - Gateway Discovery
In this section we describe the protocol for phase III.
The goal of this phase is for the active nodes from
Phase I to discover gateway connections to other
leaders that are within a hop distance of at most 3 in
Grt . During this phase, the active nodes use an aCST
range of rt.
Figure 2: Two consecutive rounds of the spanner
protocol
The active nodes use the rounds reserved in phase II to
achieve interference-free communication with the
inactive nodes within their transmission range. Each
round consists of four time slots for communication in
phase III, where each time slot represents a
communication step as shown in Figure 2. In the first
time slot, inactive nodes send CLIENT messages and in
the second time slot the active node sends a response
accordingly; in the third and fourth time slots, an
inactive node u may broadcast to its (active and
inactive) neighbors all the information it has regarding
possible gateways between the leader owning the
reserved round and other leader nodes it has heard
about. For simplicity, we assume that all active nodes
are reactivated at the same time and hence that we can
directly compare the time stamps with respect to the
different active nodes.
In reality, each inactive node u would keep
track of the offsets of the (constant number of) time
stamps it receives (in the corresponding slots allocated
to the different leaders in phase II) and use these offsets
when comparing time stamps from different leaders.
We first describe the data structures that are maintained
during this phase. Each inactive node u maintains a
cache, called Pu, which has entries of the form (L, v, tL)
where L is an active node, v is an inactive node (with u
= v possibly), and tL is the time stamp with respect to L
at which the entry (L, v) is added to Pu. When
comparing entries in the cache, a * acts as a wild card
that matches any value. The operation enqueue(L, v, tL)
on Pu is used to add the new entry (L, v, tL_) to Pu.
Enqueue performs the following checks before actually
adding the new entry to Pu. When adding a new entry
(L, v, tL_), any entry of the form (L,*, tL) with t_ < t_ is
evicted. If no such entry exists and Pu is full, then the
least recently added entry (*,*,t`), that is t_ = min{t|t <
t_ and (*,*, t) Pu}, is evicted to make room for the
new entry. The cache Pu has space enough to store a
constant, d2, number of entries. Inactive nodes also
maintain a state that is either awake or asleep with
respect to each active node that is within their
transmission range. The asleep nodes just listen the
channel and become awake when they receive a FREE
or a ACK message.
5. Conclusion
In this paper, we have examined the existing any cast
routing strategies and introduced a new class of any
cast routing schemes: density-based routing. We have
presented a field based routing model that represents
both, the existing any cast routing schemes as well as
the density-based ones. We use the results from the
model evaluation to categorize the routing strategies
into three regimes: (I) proximity-based routing; (II)
routing as the tradeoff between proximity and density;
and (III) pure density-based routing.
Our results show that density-based any cast routing is
of particular interest in wireless and mobile ad hoc
networks. Due to the dynamic nature of these networks,
traditional proximity based routing schemes often fail
to find alternative routes when a group member moves
away or when intermediate links along the path to a
group member break.
Under these conditions, density-based any cast
routing outperforms proximity-based routing in terms
of successful packet delivery because the probability to
be able to re-route packets is increased when
forwarding towards a high population of group
members. From our simulation studies we learn that the
best routing strategy lies in a tradeoff between
proximity and density, obtained using a value of k ≈ 1
in our model. This particular tradeoff offers the
increased robustness of density-based routing without
introducing a significant path stretch compared to pure
proximity-based routing. It is noteworthy that many
potential fields in physics such as the electric field or
the gravitational field follow a potential decreasing law
Divi Mydhili et al ,Int.J.Computer Technology & Applications,Vol 3 (1), 396-400
IJCTA | JAN-FEB 2012 Available [email protected]
399
ISSN:2229-6093
with a value of k = 1. It seems that physical laws can
inspire us to design better systems and algorithms.
References
[1] J. Wortham (2009, Sep.). Customers Angered as
iPhones Overload AT&T. The New York Times.
[Online]. Available: http://www.nytimes.
com/2009/09/03/technology/companies/03att.html.
[2] A. Lindgren and P. Hui, “The quest for a killer app
for opportunistic and delay-tolerant networks,” in Proc.
ACM CHANTS, 2009, pp. 59–66.
[3] D. Kim, D. Meyer, H. Kilmer, and D. Farinacci,
“Anycast Rendevous Point (RP) mechanism using
Protocol Independent Multicast (PIM) and Multicast
Source Discovery Protocol (MSDP),” RFC 3446,
January 2003.
[4] J. Moy, “OSPF Version 2,” IETF RFC 2328, April
1998.
[5] V. Park and J. Macker, “Anycast Routing for
Mobile Services,” in Conference on Information
Sciences and Systems (CISS), Baltimore, MD, USA,
March 1999.
[6] G. Malkin, “Rip version 2,” IETF RFC 2453,
November 1998.
[7] V. Park and S. Corson, Temporally-Ordered
Routing Algorithm (TORA), IETF Internet Draft, July
2001.
AUTHORS PROFILE
Divi Mydhili, Pursuing
M.Tech(CS) from Nalanda
Institute of Engineering &
Technology,Siddharth Nagar,
Sattenapalli, Guntur Affiliated to
JNTUK, Kakinada, A.P., India.
My research Interests are
computer networks.
R. Rammohan Rao, working as
Professor & Head, Department
of Computer Science
Engineering at Nalanda Institute
of Engineering &
Technology,Siddharth Nagar,
Sattenapalli, Guntur Affiliated to JNTUK, Kakinada,
A.P., India. My research Interests are Mobile
Computing, Network Security and Mobile Networks.
He is a Life member of AMIT.
Divi Mydhili et al ,Int.J.Computer Technology & Applications,Vol 3 (1), 396-400
IJCTA | JAN-FEB 2012 Available [email protected]
400
ISSN:2229-6093