Академический Документы
Профессиональный Документы
Культура Документы
8, AUGUST 2009
Multipath Dissemination in
Regular Mesh Topologies
Kranthi K. Mamidisetty, Minlan Duan, Shivakumar Sastry, Member, IEEE Computer Society, and
P.S. Sastry, Senior Member, IEEE
Abstract—Mesh topologies are important for large-scale peer-to-peer systems that use low-power transceivers. The Quality of
Service (QoS) in such systems is known to decrease as the scale increases. We present a scalable approach for dissemination that
exploits all the shortest paths between a pair of nodes and improves the QoS. Despite the presence of multiple shortest paths in a
system, we show that these paths cannot be exploited by spreading the messages over the paths in a simple round-robin manner;
nodes along one of these paths will always handle more messages than the nodes along the other paths. We characterize the set of
shortest paths between a pair of nodes in regular mesh topologies and derive rules, using this characterization, to effectively spread
the messages over all the available paths. These rules ensure that all the nodes that are at the same distance from the source handle
roughly the same number of messages. By modeling the multihop propagation in the mesh topology as a multistage queuing network,
we present simulation results from a variety of scenarios that include link failures and propagation irregularities to reflect real-world
characteristics. Our method achieves improved QoS in all these scenarios.
Index Terms—Wireless communication, network communications, packet-switching networks, routing protocols, mesh topology.
1 INTRODUCTION
such identifiers. Several techniques, called dissemination
T HE use of novel devices for computing and communica-
tion in highly engineered networked embedded systems
such as streetlight management [1], reconfigurable conveyors
methods, were developed at the network layer to regulate
the flow of messages between nonadjacent nodes without
[2], and critical infrastructures [3], presents new challenges relying on unique node identifiers or constructing routing
and opportunities for monitoring and diagnostics. These tables using these identifiers [7], [8], [9].
systems contain a large number of nodes, with each node In this paper, we consider highly engineered systems
incorporating a tiny microcontroller, sensors, actuators, and comprising of nodes arranged in a regular mesh topology.
an integrated low-power transceiver [4]. The nodes interact in We focus on methods for effectively utilizing all the shortest
a peer-to-peer manner over low-bandwidth wireless links to paths available between a pair of nodes and present results
achieve the application objectives. To increase the number of to show that effective utilization of all the available paths
simultaneous interactions between the nodes in the system, significantly improves the Quality of Service (QoS).
the transmission range of each node is limited so that it In many highly engineered systems, one can assume that
communicates directly only with its set of immediate the nodes have fixed relative locations. Often, the systems are
neighbors; such an arrangement of nodes is referred to as a designed to overlay on an underlying grid. For example, in
mesh topology [5]. automation systems, the regions demarcated by such a grid
Multihop communications are necessary in such systems are called zones [10]; a zone is a commonly used abstraction to
to send messages from any source to any destination. For support the design, operation, and maintenance activities.
example, intermediate nodes must forward messages to a Motivated by applications in such domains, we consider
monitoring station from nodes that cannot communicate regular mesh topologies that arise by embedding the nodes in
directly with the monitoring station. Routing protocols are a 2D Basegrid. We show in Section 2 that several mesh
used extensively in wired and wireless networks to support topologies arise when the location in the 2D Basegrid and the
transmission range of the nodes change. Because the grid
multihop communication [6]. Such protocols construct and
coordinates can specify the nodes, the shortest paths between
maintain routing tables at each node by relying on system- any pair of nodes can be locally computed. Most routing
wide unique node identifiers. When the number of nodes is protocols select only one of the shortest paths even when
very large, such as in sensor networks, it is not feasible to use multiple such paths exist. This results in reduced system-
level QoS [11]. We address the issue of how to effectively
utilize all the shortest paths available. Since the resulting
. K.K. Mamidisetty, M. Duan, and S. Sastry are with the Department
of Electrical and Computer Engineering, University of Akron, Akron,
methods amount to a node making local decisions on how to
OH 44325-3904. distribute messages among its immediate neighbors, without
E-mail: {km50, ssastry}@uakron.edu, MINLAN_DUAN@denso-diam.com. having to dynamically construct any routing tables, we refer
. P.S. Sastry is with the Department of Electrical Engineering, Indian to this method of forwarding messages as dissemination in
Institute of Science, Bangalore 560 012, India. spite of the fact that nodes are identified by their global
E-mail: sastry@ee.iisc.ernet.in.
coordinates in the underlying 2D Basegrid.
Manuscript received 11 Sept. 2007; revised 10 Apr. 2008; accepted 11 July Because each node communicates directly with its im-
2008; published online 27 Aug. 2008. mediate set of neighbors, there are multiple shortest paths
Recommended for acceptance by T. Abdelzaher.
For information on obtaining reprints of this article, please send e-mail to: between many pairs of nodes in a mesh topology. The number
tpds@computer.org, and reference IEEECS Log Number TPDS-2007-09-0312. of such paths is limited by the relative locations of the nodes.
Digital Object Identifier no. 10.1109/TPDS.2008.164. For example, the number of shortest paths between certain
1045-9219/09/$25.00 ß 2009 IEEE Published by the IEEE Computer Society
MAMIDISETTY ET AL.: MULTIPATH DISSEMINATION IN REGULAR MESH TOPOLOGIES 1189
pairs of nodes is one, despite the mesh topology. We define a discusses contours, their structure, and the significance of
Contour as the union of all the shortest paths between a pair of the structure on dissemination methods. Section 4 presents
nodes and present some results to precisely characterize the results to show that when the messages are spread in a
structure of contours. Using this structure, we show that round-robin manner, nodes along one path in the contour
when messages are spread in a round-robin manner, nodes will always handle more messages than nodes along other
along one path in the contour will always handle more paths. Section 5 presents optimal spreading rules that
messages than the nodes along other paths in the contour. ensure that all the nodes, which are at a same distance from
Consequently, the benefits of the multiple paths cannot be the source in a contour, handle roughly the same number of
fully realized. We then present a strategy for spreading messages. After presenting simulation results to illustrate
messages to neighboring nodes that effectively exploits the the QoS of CGD in Section 6, we conclude in Section 7.1
available shortest paths and show that our rules for spreading
the messages result in a balanced loading of all the available
shortest paths. We refer to this approach as Contour Guided 2 EMBEDDING FUNCTIONS
Dissemination (CGD). We consider mesh topologies that are obtained by embed-
CGD improves the QoS by disseminating the messages ding nodes on a 2D Basegrid and adjusting the transmission
over all the available shortest paths. Typical application range of each node. Each location on a 2D Basegrid is
scenarios in which CGD will be useful are: 1) a node identified by a unique ordered pair ði; jÞ. The distance
responding to a diagnostic query from an operator at a
between two consecutive locations on the grid, i.e., between
monitoring station, and 2) a node sending a recorded incident
ði; jÞ and ði; j þ 1Þ or between ði; jÞ and ði þ 1; jÞ, is b.
to a monitoring station in a surveillance application. CGD
Let N denote a set of nodes. An embedding function
equitably disseminates the messages over all the available
shortest paths in a manner that maximizes utilization of all assigns a location on the 2D Basegrid and a transmission
available shortest paths. We present simulation results to range to each node, ni 2 N :
demonstrate all these aspects. : ni 2 N ! ðIN IN IRÞ: ð1Þ
Routing protocols used in traditional wired and wireless
networks are based on shortest path algorithms such as the The location to which a node is assigned is assumed to be
Bellman-Ford algorithm [12] and Dijkstra’s algorithm [6]. fixed for the application. Each node knows its own location
Similar protocols have been reported for ad hoc, wireless, and the number of neighbors it has. Further, we assume that
and mobile networks [13], [14], [15], [16]. The QoS achieved all the neighbors of a node reliably receive messages sent by
in these systems has also been studied [17], [18], [19]. The the node; however, in our simulations, we empirically
dissemination method we describe in this paper is some- explore robustness to temporary link failures. Each message
what similar to a gradient dissemination scheme [13] with contains the locations of the source and destination nodes as
the cost metric being the deviation from evenness of load specified by the embedding function.
distribution on all available shortest paths. Different mesh topologies are obtained by specifying
Recent efforts have focused on exploiting multipaths to different embedding functions. Suppose that the nodes in N
improve the QoS in systems [7], [20], [21] using constrained
must be embedded along R rows and C columns, starting at
node distribution models. For example, in [21], the nodes are
location ðX; Y Þ of the 2D Basegrid.2 Let q denote the
distributed randomly in a unit disk. In [7], the nodes are
embedding in which each node n 2 N communicates directly
distributed in a narrow strip no wider than 0.86 times the
transmission range of a node. The multipath, multihop, with q neighbors. For q ¼ 4, the embedding function
approach we present in this paper also assumes a constrained
i1
distribution of the nodes, namely, nodes are on a 2D Basegrid. 4 : ni 2 N ! X þ ði 1Þ mod C; Y þ ;b ð2Þ
While there is no stochasticity in the node distribution, we C
precisely characterize the geometry of the shortest paths and yields a mesh topology. Since the transmission range for
show how to exploit it to achieve better QoS. each node is b, it can only communicate with its four
In the current methods, the motivating factors for
immediate neighbors. Similarly, for q ¼ 8, the embedding
considering multipath routing include fault tolerance, high-
function
er aggregate bandwidth, and load balancing [22], [20], [23],
[24], [25]. The QoS aspects of multipath routing have been
i 1 pffiffiffi
addressed in [26], [27], [28], and [29]. The split multipath 8 : ni 2 N ! X þ ði 1Þ mod C; Y þ ; b 2 ; ð3Þ
C
routing protocol maintains maximally edge-disjoint paths pffiffiffi
[30]. Braided multipaths discussed in [31] are useful when where the transmission range is b 2, yields the overlapping
the routing is coupled with diagnostic or prognostic mesh topology shown in Fig. 1a. Similarly, other embed-
methods to select alternative paths. All such methods dings such as 6 shown in Fig. 1b arise by changing the
essentially focus on the discovery and maintenance of location and transmission radius of the nodes [33].
multiple paths that are useful under various constraints on In the remainder of this paper, nodes that are assigned to
the node distribution. In contrast, we assume regular mesh location ði; jÞ on the 2D Basegrid are referred to as ni;j . In
topologies and then precisely characterize the set of all the next section, we define contours and their structure
shortest paths between any pair of nodes. We then use this precisely. The results for contours in the 8 embedding are
geometric structure to propose rules for dissemination that
presented in detail. Contours in other embeddings are
result in all available shortest paths being utilized effec-
briefly discussed at the end of the section.
tively so that QoS improves.
The rest of the paper is organized as follows: Section 2 1. A preliminary version of this work was reported in [32].
presents embedding functions that characterize a class of 2. Our method depends only on the relative locations of the source and
mesh topologies that we consider for this work. Section 3 destination nodes, and hence, the specific ðX; Y Þ is not important.
1190 IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 20, NO. 8, AUGUST 2009
Lemma 1. In 8 , the contour of ni;j and nq;r is a single shortest Because y is odd, the length of the path from nq;j to
path if and only if xy ¼ 0. nqrþj;j must also be odd. Hence, the length of the path
from ni;j to nqrþj;j is even because x is odd. It follows
Proof. When xy ¼ 0, we know that q i ¼ r j. The path
from Lemma 3 that the contour of ni;j and nqrþj;j is a
< ni;j ; niþ1;jþ1 ; . . . ; nq1;r1 ; nq;r > is a shortest path in the rectangle with ¼ bqrþji c and is enclosed by the
2
embedding 8 between ni;j and nq;r . All other nodes ns;t corner nodes: ni;j , niþ;j , niþ;jþ , and nqrþj;j . Let us
violate the criterion in (9) and are hence are not in the refer to this contour as B. Since the four corner nodes
contour. of B satisfy the distance condition in (9), we can
Next, we need to show that xy ¼ 0 when the conclude that all the nodes in contour B are also in the
contour is a single path. Suppose xy 6¼ 0, and without contour of ni;j and nq;r .
loss of generality, let x > y . By our notation, this Now, consider the node niþrj;r . By the criterion of (9),
means that the length of the shortest path is q i. Let this node would be in the contour, and no node ns;r with
K > 0 be such that qi ¼ rjþK. Using (8), by direct s < i þ r j would be in the contour. We can see that
calculation, we have dðni;j ; niþK;j ÞþdðniþK;j ; nq;r Þ ¼ ni;j , niþ1;jþ1 ; . . . ; niþrj;r is the unique shortest path
dðni;j ; nqK;r Þ þ dðnqK;r ; nq;r Þ ¼ dðni;j ; nq;r Þ. This implies between ni;j and niþrj;r . Since y is odd, the length of
that nodes nqK;r and niþK;j have to be on optimal this shortest path must be odd. Consequently, the length
paths. However, using the distance criterion (9), it is of the shortest path from niþrj;r to nq;r via the nodes
easily seen that a path from ni;j to nq;r through both niþrjþ1;r ; niþrjþ2;r ; . . . ; nq1;r ; nq;r must be even. From
these nodes is not a shortest path. This implies we Lemma 3, the contour of niþrj;r and nq;r is a rectangle
have to have at least two shortest paths. But since we with ¼ bqrþji c and is enclosed by the nodes nq;r ,
2
are given the fact that the contour is a single path, we nq;r , nqþ;r , and nq;r . Let us refer to this contour as
must have K ¼ 0, which is same as xy ¼ 0. u
t T . Since the four corner nodes of T also satisfy the
Lemma 2. In 8 , the contour of ni;j and nq;r comprises two distance condition in (9), we conclude that all the nodes
shortest paths that are enclosed by the nodes ni;j , niþ1;j , nq1;r , in contour T are also in the contour of ni;j and nq;r .
and nq;r if and only if xy ¼ 1. Consider the path < ni;j ; niþ1;jþ1 ; niþ2;jþ2 ; . . . ; niþrjþ;
r þ > . Let us refer to this path as line-L. From the
Proof. Given the nodes as stated, we can use the distance definition of (7), i þ r j þ ¼ q , and hence, line-L
criterion in (9) to verify that node niþ1;j and node nq1;r starts at node ni;j and ends at nq;rþ . Similarly, let us
are on different shortest paths from ni;j to nq;r . It now refer to the path < nq;r ; nq1;r1 ; nq2;r2 ; . . . ; niþ;j > as
suffices to note that nodes niþ2;j and nq2;r do not satisfy line-R. Every node ns;t that is not in contour T or
the distance criterion in (9) and, hence, these are the only contour B which is in the contour of ni;j and nq;r must lie
two shortest paths possible. between line-L and line-R. Any node not in the contours
Given two shortest paths between ni;j and nq;r , we now T and B, which is not between line-L and line-R, does not
show that the paths are enclosed by the four nodes as satisfy the distance criterion in (9). We can therefore
stated. We know that a path from nq;r to ni;j is a shortest conclude that the contour of ni;j and nq;r is a rectangle
path only if maxfx ; y g is reduced along each step of the with line-L and line-R as two edges, and hence, the four
path. Since by our notation, q i > r j, x must reduce corner nodes are at either ends of these lines.
along each step of the path. In the 8 embedding, there are Next, given a contour of ni;j and nq;r with four corner
only three neighbors of nq;r with a lower x , namely, nodes as stated in the theorem, we show that xy ðni;j ; nq;r Þ
nq1;r1 , nq1;r , and nq1;rþ1 . nq1;rþ1 does not satisfy the
distance criterion in (9). Using similar arguments in the is even. xy ðni;j ; nq;r Þ is even only when x ðni;j ; nq;r Þ and
neighborhood of ni;j , we conclude that only nodes niþ1;jþ1 y ðni;j ; nq;r Þ are both even or are both odd. By definition,
and niþ1;j can be on any shortest path between ni;j and nq;r . ðni;j ; nq;r Þ ¼ jqijjrjj
2 , and ðni;j ; nqrþj;j Þ ¼ jðqrþjÞijjjjj
2 .
Finally, we note that only the nodes on the two shortest Thus, ðni;j ; nq;r Þ ¼ ðni;j ; nqrþj;j Þ. niþ;j is a corner
paths satisfy the distance criterion in (9), and hence, the node in the contour of ni;j and nqrþj;j . x ðni;j ; nqrþj;r Þ ¼
four enclosing nodes are as stated. u
t
2 ðni;j ; nqrþj;j Þ by the structure of the 2D BaseGrid and
Lemma 3. In 8 , given ni;j and nq;r with y ¼ 0, the contour is a hence is even by Lemma 3. Since xy ðni;j ; nqrþj;r Þ ¼ 0, the
rectangle if and only if x is even. The four nodes at the distance dðnqrþj;j ; nq;j Þ is always equal to dðnq;r ; nq;j Þ.
corners of this rectangle are ni;j , niþ;jþ , nq;j , and nq;j .
Therefore, x ðni;j ; nq;r Þ and y ðni;j ; nq;r Þ are either both
(The proof is available in [32].)
even or both odd. Hence, xy ðni;j ; nq;r Þ must be even. t u
Theorem 1. In 8 , the contour of ni;j and nq;r is a rectangle
Lemma 4. In 8 , given ni;j and nq;r with y ¼ 0, the contour is
bounded by the nodes ni;j , niþ;j , nq;rþ , and nq;r if and
enclosed by a hexagon if and only if x is odd and x > 1. The
only if xy is even and xy > 0. six nodes at the corners of this hexagon are ni;j ; niþ; j ;
Proof. Given that xy is even, we first show that the corner niþþ1; j ; nq;jþ ; nq1;jþ and nq;j . (The proof is avail-
nodes of the contour are as stated. Since xy is even, able in [32].)
either x and y are both even or they are both odd. Let Theorem 2. In 8 , the contour of ni;j and nq;r is a hexagon
x and y be both odd without loss of generality. bounded by the nodes ni;j , niþ;j , niþþ1;j , nq;rþ ,
Consider the node nqrþj;j . By the criterion of (9),
nq1;rþ , and nq;r if and only if xy is odd and xy > 1.
this node would be in the contour, and no node ns;j
with s > q r þ j would be in the contour. The only Proof. The proof is similar to that of Theorem 1 by using
shortest path from node nqrþj;j to nq;r is through the Lemma 4 instead of Lemma 3. u
t
nodes nqrþjþ1;jþ1 ; nqrþjþ2;jþ2 ; . . . ; nq1;r1 (by Lemma 1).
1192 IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 20, NO. 8, AUGUST 2009
5 OPTIMAL SPREADING
We now present an algorithm for spreading the messages so
that all the available paths are effectively utilized. Recall
that a row is a collection of nodes in the contour that are at
the same distance from the source. Let w be the number of
nodes in a row of a contour. We refer to w as the width of
the row. If the source sends M messages and if every node
5
in every row handles M w messages, then we can say that the
spreading is the best in the sense that all available paths are
effectively used. This is the criterion of optimality that we
choose. We will show that the algorithm presented in this
section is optimal in this sense.
Fig. 6 shows examples of two contours—one in 8 and the
other in 4 . Notice that the width of a row, w, varies with the
row number in three ways in both the contours. w first
increases monotonically, then remains constant, and then
decreases monotonically. We refer to these three regions as
the Expansion region, Propagation region, and Contraction
region, respectively.
Observation 2. In the 8 embedding, the width w of any row in the
expansion region is odd, and the width of the next downstream
row is w þ 2. When xy > 0 and even, the width of the rows in
the propagation region is the same as the largest w in the
expansion region; when xy > 0 and odd, the width of the rows in
Fig. 6. Expansion, propagation, and contraction regions. the propagation region is w þ 1. The width of each successive row
(a) Contour in 8 . (b) Contour in 4 .
in the contraction region reduces by two until the ni;j is reached.
Consider two adjacent rows m and m þ 1 in the
Proof. First, consider nodes along Path-2. Since each node
expansion region with widths w and w þ 2, respectively.
sends the messages it handles to three downstream nodes,
Since there are more nodes in row m þ 1, if all of them have
after m ¼ log3 M steps on Path-2, the node nqm;rm will
to handle the same number of messages, then at least some
handle at most one message. This node must be in row m of
of them have to receive messages from two nodes in row m.
the contour. From the properties of trinomial coefficients,
The strategy we propose uses a special labeling of the nodes
the node nqm;rmþ1 , on Path-ð2 1Þ path, will handle
in a row; every node can compute its own label using the
m messages. After m0 ¼ log3 log3 M more steps along Path-
contour structure results discussed in Section 3.
ð2 1Þ, node nqðmþm0 Þ;rðmþm0 Þþ1 will handle at most
m0 messages. Extending this argument, we conclude that 5.1 Node Labeling
in Path-k, 2; . . . ; 2, some node nqOðlog3 MÞ; rOðlog3 MÞþ2k In any row, nodes on both sides of the middle node are labeled
must handle at most one message. Since the messages as 1; 2; 3; . . . ; bw=2c, as illustrated in Fig. 7. Let nmp represent the
always move from nodes in higher numbered layers to
node that is labeled p in the mth row of a contour. Note that nm p
nodes in lower numbered layers, such messages must
is not the ID (or coordinates) of the node; in each row, two
eventually arrive at the nodes on the Layer-0 path. Hence,
nodes with different IDs would have the same label. However,
we have the theorem. u
t
for any node except the middle one in row m, only one of the
4.2 Effects of Embedding on Loading two nodes with label p would be a neighbor. Thus, for all
We observed similar loading effects under uniform spread- nodes except the middle one in row m, the node label as per the
ing in 4 and 6 . Recall that the structure of the contours in convention shown in Fig. 7 uniquely identifies a neighbor in
6 closely follows that in 4 . In these embeddings, only the row m þ 1. Under this labeling in the 8 embedding, node nm p
nodes along paths on either the X or Y axis get loaded. If mþ1
can communicate with nodes npmþ1 and npþ1 . For the middle
x > y , then the path along the X-axis in the contour will node in row m, there would be two neighbors in row m þ 1
be loaded. Proofs of loading, similar to the ones presented with the same label.
for 8 in the preceding section, for contours in 4 and 6
follow from the properties of binomial coefficients. 5. This is only possible to the extent that M is an integral multiple of w.
MAMIDISETTY ET AL.: MULTIPATH DISSEMINATION IN REGULAR MESH TOPOLOGIES 1195
5.2 Optimal Spreading ensure that all the nodes in every row of a contour handle
Our algorithm for optimal spreading is given below. In roughly the same number of messages. Theorem 4 below
the algorithm, the array ngbrs½ keeps track of the relevant shows that this objective is achieved when each node in the
neighbors for any node. As explained above, the middle expansion region and the contraction region spread messages
node in a row has two relevant neighbors in the next row to its neighbors in the ratios specified in the Optimal
that have the same label. Spreading algorithm. The empirical results in the next section
Optimal Spreading Algorithm show that this notion of optimality significantly improves the
QoS.
Theorem 4. When the source in a contour sends M messages and
there is no message loss, under the Optimal Spreading algorithm,
6
every node in each row of width w handles M w messages.
Proof. We prove the result for the expansion region of the
contour using mathematical induction. As a basis, n11 is in
a row with width w ¼ 1, and it handles M w ¼ M messages.
This node sends messages to three nodes in the next row
in a specific ratio, as given in the algorithm. The middle
M
node in row 2 receives wþ2 ¼ M3 messages from n11 . The
2
two nodes labeled n1 (recall that the middle node in row
m has two neighbors with the same label in row m þ 1)
M
receive wðwþ2Þ ¼ M3 messages each. Thus, all the nodes in
row 2 that has width w ¼ 3 handle M3 messages.
Assume that the nodes in row m of the contour with
width w handle M w messages. We show that all the nodes
M
in row m þ 1 handle wþ2 messages. As per our algorithm,
mþ1
node np sends messages to npmþ1 and npþ1
m
in the ratio
M
w þ 2 2p : 2p. Since this node has w messages, this ratio
Mp Mðp1Þ
implies that nmp sends wþ2 w messages to npmþ1 and
Mp Mp mþ1
w wþ2 messages to npþ1 . Following this rule, the
ð11ÞM
nodes n1mþ1 receive 1M wþ2 w
M
¼ wþ2 messages from
m
the nodes n1 .
The nodes njmþ1 , 2 j ðbwþ2 2 c 1Þ, receive
. ðj1ÞM
w ðj1ÞM
wþ2 messages from nodes nm j1 , and
jM ðj1ÞM
. wþ2 w messages from nodes nmj .
Hence, each node nmþ1 j
M
handles wþ2 messages.
Now, consider nodes nbmþ1 1
wþ2 . These nodes receive 2
2 c
ðM M
w wþ2Þ messages from the middle node in row m and
bw=2cM
w bw=2cM
wþ2 messages from nodes nm bw=2c . Thus,
mþ1 M
nodes nbðwþ2Þ=2c also handle wþ2 messages. Therefore,
M
since the middle node in row m þ 1 receives wþ2
messages from the middle node in row m, we conclude
that the spreading rules presented are optimal for the
expansion region.
Our algorithm for spreading messages is based on the It is easy to see that if the number of messages
following strategy: handled by the nodes in a row at the beginning of the
propagation region are roughly the same, this number
. In the expansion region, nodes nm p send messages to will not change since the nodes do not spread messages.
nodes nmþ1
p and nodes nmþ1
pþ1 in different ratios. The proof for the contraction region also follows from
. In the propagation region, node nm p sends all the similar induction, and the details are omitted. u
t
messages it handles to node npmþ1 . In the above proof (and in specifying our algorithm), it is
. In the contraction region, nodes nm p send messages assumed that the width of rows in the propagation region is
to node npmþ1 and nodes nmþ1
p1 in different ratios, as the same as the largest width in the expansion region.
shown in the algorithm. When the contour is a hexagon (i.e., when xy > 0 and
Intuitively, the nodes in the expansion region spread odd), there are two special cases that must be considered
messages over the available paths, the nodes in the propaga- because if w is the maximum width of the expansion region,
tion region do not spread messages and propagate them the width of the propagation region is w þ 1, and the
along a single path, and the nodes in the contraction region
coalesce messages from multiple paths. Our objective is to 6. This is only possible to the extent that M is an integral multiple of w.
1196 IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 20, NO. 8, AUGUST 2009
6 RESULTS
To evaluate the QoS achieved by CGD, we designed and
implemented a discrete-event simulation using the OM-
Net++ framework [34]. The results in this section demon-
strate the performance of CGD. For all simulations, the nodes
were embedded on the 2D Basegrid using the 8 embedding
with xy ¼ 4, one source, and one sink in the contour. In the
scenarios where the number of messages was not varied, the
source sent 5,000 messages, each of which is 36 bytes long. To
study the effects of an increased number of messages, the
number was changed between 5 and 50,000.
in Fig. 8b, all the nodes in a row handle roughly the same
number of messages in optimal CGD.
Fig. 11. The average latency and jitter achieved using optimal CGD is Fig. 12. When the nodes with irregular wireless propagation can be
significantly lower even when a large number of nodes in the contour localized to the expansion region of a contour, the message loss rate
have irregular propagation. (a) Average latency. (b) Jitter. improves. (a) Average latency. (b) Message loss rate.
6.4 Summary and Discussion Fig. 13. The average latency and throughput achieved using optimal
The preceding results demonstrate that despite the avail- CGD are significantly better than those of uniform CGD. (a) Average
ability of multiple shortest paths between a pair of nodes, latency. (b) Throughput.
MAMIDISETTY ET AL.: MULTIPATH DISSEMINATION IN REGULAR MESH TOPOLOGIES 1199
7 CONCLUSIONS
Many future engineered systems that are based on peer-to-
peer-connected mesh topologies are likely to have multiple
paths between a pair of nodes. We defined a contour as the
union of all shortest paths between a pair of nodes. Using a
regular topology, we proved that when the messages are
spread uniformly over the paths in a contour, nodes along
one path handle more messages than other messages. We
presented an optimal strategy for spreading messages in
such systems, and our results demonstrate the effectiveness
of the spreading strategy.
Although the results are based on regular topologies,
they represent upper bounds on what can be achieved in
general topologies. These results reveal that to achieve Fig. 16. Layered paths when x is even.
1200 IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 20, NO. 8, AUGUST 2009
The node ni;jþ1 cannot be on any shortest path because [10] J. Agre, L. Clare, and S. Sastry, “A Taxonomy for Distributed
any path between ni;j and nq;j that passes through ni;jþ1 Real-Time Control Systems,” Advances in Computers, vol. 49,
pp. 303-352, 1999.
will have length x þ 1. Similarly, node nq;jþ1 cannot [11] K. Russell, “Maximizing Performance for Wireless Mesh Net-
be on any shortest path between ni;j and nq;j . works,” EE Times, Feb. 2007.
In each layer k, 1 k , the number of nodes that [12] R.E. Bellman, Dynamic Programming. Princeton Univ. Press, 1957.
can be on any shortest path between ni;j and nq;j [13] K. Akkaya and M. Younis, “A Survey on Routing Protocols for
reduces by 2 k. For example, the Layer-1 path will Wireless Sensor Networks,” Ad Hoc Networks, vol. 3, pp. 325-349,
2005.
have ðx þ 1Þ 2 nodes, the Layer-2 path will have [14] D.B. Johnson and D.A. Maltz, “Dynamic Source Routing in Ad Hoc
ðx þ 1Þ 4 nodes, and so on. Hence, the Layer- path Wireless Networks,” Mobile Computing, Imielinski and Korth, eds.,
must have a single node, i.e., niþ;jþ since the Layer-0 vol. 353, pp. 153-181, Kluwer Academic Publishers, 1996.
path has an odd number of nodes because x is even. [15] S. Murthy and J.J. Garcia-Luna-Aceves, “An Efficient Routing
Protocol for Wireless Networks,” Mobile Networks and Applications,
Similar arguments apply to the Layer -1, Layer -2, . . . , vol. 1, no. 2, pp. 183-197, 1996.
Layer- paths, and the Layer- path must have a single [16] C.E. Perkins and P. Bhagwat, “Highly Dynamic Destination-
node, i.e., nq;j . Hence, the four corners of the Sequenced Distance-Vector Routing (DSDV) for Mobile Compu-
rectangle are as required. ters,” Proc. ACM SIGCOMM ’94, pp. 234-244, 1994.
Next, given a contour with four corner nodes ni;j , [17] D. Chen and P.K. Varshney, “QoS Support in Wireless Sensor
Networks: A Survey,” Proc. Int’l Conf. Wireless Networks
niþ;jþ , nq;j , and nq;j , we show that x is even. (IWCN ’04), June 2004.
Consider the path from ni;j to niþ;jþ . The length of this [18] T. He, J.A. Stankovic, C. Lu, and T. Abdelzaher, “SPEED: A
path is the same as the length of the path from ni;j to Stateless Protocol for Real-Time Communication in Sensor Net-
niþ;j because of the structure of the 8 embedding. x is works,” Proc. 23rd Int’l Conf. Distributed Computing Systems
2 lengthðni;j ; niþ;j Þ and is hence even. u
t (ICDCS), 2003.
[19] C. Lu, B.M. Blum, T.F. Abdelzaher, J.A. Stankovic, and T. He,
Proof (Lemma 4). The arguments are similar to the ones “RAP: A Real-Time Communication Architecture for Large-Scale
Wireless Sensor Networks,” Proc. Eighth IEEE Real-Time and
used in the proof of Lemma 3. The only difference is that
Embedded Technology and Applications Symp. (RTAS), 2002.
the Layer- path will have two nodes instead of one [20] S. Muller, R.P. Tsang, and D. Ghosal, “Multipath Routing in
because there are an even number of nodes in the Layer-0 Mobile Ad Hoc Networks: Issues and Challenges,” Lecture Notes in
Computer Science, Springer, 2004.
path, and the number of nodes on each successive layer [21] E. Hyytia and J. Virtamo, “On Traffic Load Distribution and Load
reduces by two. u
t Balancing in Dense Wireless Multihop Networks,” EURASIP
J. Wireless Comm. and Networking, vol. 2007, pp. 1-15, 2007.
[22] R. Draves, J. Padhye, and B. Zill, “Routing in Multi-Radio,
ACKNOWLEDGMENTS Multi-Hop Wireless Mesh Networks,” Proc. ACM MobiCom,
Sept./Oct. 2004.
The authors thank the anonymous reviewers whose com- [23] P.P. Pham and S. Perreau, “Performance Analysis of Reactive
ments on an earlier draft of this paper have helped to Shortest Path and Multi-Path Routing Mechanism with Load
Balance,” Proc. IEEE INFOCOM, 2003.
improve the presentation. This material is based upon work [24] A. Valera, W. Seah, and S.V. Rao, “Cooperative Packet Caching
supported by the US National Science Foundation (NSF) and Shortest Multipath Routing in Mobile Ad Hoc Networks,”
Proc. IEEE INFOCOM, 2003.
under Grant 0720736. Any opinions, findings, and conclu-
[25] L. Wang, Y. Shu, M. Dong, L. Zhang, and W.W.O. Yang,
sions or recommendations expressed in this material are “Adaptive Multipath Source Routing in Ad Hoc Networks,” Proc.
those of the authors and do not necessarily reflect the views IEEE Int’l Conf. Comm. (ICC ’01), vol. 3, 2001.
[26] Y.S. Chen, C.Y. Tseng, J.P. Shue, and P.H. Kuo, “An On-Demand,
of the NSF. Link-State, Multi-Path QoS Routing in a Wireless Mobile Ad-Hoc
Network,” Proc. European Wireless Conf. (EW), 2002.
[27] A. Fallahi, E. Hossain, and A.S. Alfa, “QoS and Energy Trade
REFERENCES Off in Distributed Energy-Limited Mesh/Relay Networks: A
[1] R.N. Murthy, A. Gosain, M. Tierney, A. Brody, A. Fahad, J. Bers, Queuing Analysis,” IEEE Trans. Parallel and Distributed Systems,
and M. Welsh, “CitySense: A Vision for an Urbanscale Wireless vol. 17, no. 6, pp. 576-592, June 2006.
Networking Testbed,” Proc. ACM Sixth Workshop Hot Topics in [28] R. Leung, J. Liu, E. Poon, A. Chan, and B. Li, “MP-DSR: A QoS-
Networks (HotNets ’07), Nov. 2007. Aware Multi-Path Dynamic Source Routing Protocol for Wireless
[2] N. Hayslip, S. Sastry, and J. Gerhardt, “Networked Embedded Ad-Hoc Networks,” Proc. 26th Ann. IEEE Conf. Local Computer
Automation,” Assembly Automation, vol. 26, no. 3, pp. 235-241, 2006. Networks (LCN ’01), pp. 132-141, 2001.
[3] J.Z. Hecker, “PORT SECURITY—Nation Faces Formidable Chal- [29] W.H. Liao, C.Y. Tseng, S.L. Wang, and J.P. Sheu, “A Multi-Path
lenges in Making New Initiatives Successful,” technical report, US QoS Routing Protocol in a Wireless Mobile Ad-Hoc Network,”
General Accounting Office, 2002. Proc. First Int’l Conf. Networking (ICN), 2001.
[4] J. Hill, M. Horton, R. Kling, and L. Krishnamurthy, “The Platforms [30] S.J. Lee and M. Gerla, “Split Multipath Routing with Maximally
Enabling Wireless Sensor Networks,” Comm. ACM, vol. 47, no. 6, Disjoint Paths in Ad Hoc Networks,” Proc. IEEE Int’l Conf. Comm.
pp. 41-46, June 2004. (ICC ’01), vol. 10, 2001.
[5] I.F. Akyldiz, X. Wang, and W. Wang, “Wireless Mesh Networks: [31] D. Ganesan, R. Govindan, S. Shenker, and D. Estrin, “Highly-
A Survey,” Computer Networks, vol. 47, pp. 445-487, 2005. Resilient, Energy-Efficient Multipath Routing in Wireless
[6] G. Coulouris, J. Dollimore, and T. Kindberg, Distributed Systems Sensor Networks,” Mobile Computing and Comm. Rev., vol. 1,
Concepts and Design, fourth ed. Addison-Wesley, 2005. no. 2, 2002.
[7] J. Gao and L. Zhang, “Load-Balanced Short-Path Routing in [32] I.-H. Chu, M. Duan, and S. Sastry, “Contour Guided Dissemina-
Wireless Networks,” IEEE Trans. Parallel and Distributed Systems, tion for Networked Embedded Systems,” Proc. Int’l Symp.
vol. 17, no. 4, pp. 377-388, Apr. 2006. Innovative Real-Time Applications of Distributed Sensor Network
[8] C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann, and (IRADSN ’06), Oct. 2006.
F. Silva, “Directed Diffusion for Wireless Sensor Networking,” [33] K.K. Mamidisetty, “Generalizing Contour Guided Dissemina-
IEEE/ACM Trans. Networking, vol. 11, no. 1, pp. 2-16, 2003. tion in Mesh Topologies,” MS thesis, Univ. of Akron, May
[9] J. Kulik, W. Heinzelman, and H. Balakrishnan, “Negotiation- 2008.
Based Protocols for Disseminating Information in Wireless Sensor [34] A. Vargas, “Omnet++Discrete Event Simulation System,” Proc.
Networks,” Wireless Networks, vol. 8, pp. 169-185, 2002. European Simulation Multiconf., pp. 319-324, June 2001.
MAMIDISETTY ET AL.: MULTIPATH DISSEMINATION IN REGULAR MESH TOPOLOGIES 1201
[35] G. Zhao, T. He, S. Krishnamurthy, and J.A. Stankovic, “Impact Shivakumar Sastry received the bachelor’s
of Radio Irregularity on Wireless Sensor Networks,” Proc. Second degree in electronics engineering from Banga-
Int’l Conf. Mobile Systems, Applications, and Services (MobiSys ’04), lore University in 1984, the master’s degree in
pp. 125-138, 2004. electrical engineering from Indian Institute of
Science, the master’s degree in computer
Kranthi K. Mamidisetty received the bachelor’s science from the University of Central Florida,
degree in electronics and communication en- and the PhD degree from Case Western Reserve
gineering from Jawaharlal Nehru Technological University. Prior to his transition into academics,
University and the master’s degree in electrical he held industry positions with GE Consulting
engineering from the University of Akron, Ohio, Services and Rockwell Automation. Since 2002,
in 2008. He is currently pursuing his PhD degree he has been a member of the faculty in the Department of Electrical and
at the Department of Electrical and Computer Computer Engineering, University of Akron, Akron, Ohio. He is the
Engineering, University of Akron. recipient of several awards and holds two US patents. He is a member of
the IEEE Computer Society.