2005 Conference on Information Sciences and Systems, The Johns Hopkins University, March 1618, 2005

Maximum throughput scheduling in time-varying-topology wireless ad-hoc networks

Anna Pantelidou, Anthony Ephremides, Andre Tits
Institute for Systems Research
Department of Electrical and Computer Engineering
University of Maryland College Park, MD 20742
e-mail: {apantel, etony, andre}@umd.edu
Abstract In wireless ad-hoc networks, signal interference and
collisions from simultaneous transmissions by neighboring nodes
significantly degrade throughput. Hence, when throughput is of the
essence, it is necessary to devise scheduling policies for coordinating
the wireless transmissions. In this paper, we focus on the problem
of scheduling the transmissions of a finite number of nodes in a mobile ad-hoc wireless network. As a consequence of node mobility,
the topology varies with time. This is modeled as an i.i.d topology process. Customers of different classes enter the network and
may be serviced differently, according to their class requirements.
We propose a family of scheduling policies all of which maximize
the total stable throughput of the network, while taking into account the different class requirements. Policies within this family
are parameterized by the preferential weights assigned to the respective customer classes. Finally, we report results obtained by a
Monte Carlo simulation that confirm that mobility may considerably improve the network throughput under our proposed scheduling scheme.
Keywords: wireless networks, scheduling, maximum throughput

I. I NTRODUCTION
One of the fundamental characteristics of wireless ad-hoc networks
is the fact that the wireless medium is shared by multiple static or mobile
nodes. As simultaneous transmissions by neighboring nodes interfere
with each other and the transmitted signals get corrupted, the throughput
of the network degrades significantly.
For this reason, medium access control (MAC) schemes coordinate
the access of the various wireless nodes to the medium. Two major medium access control schemes have been widely used. On the
one hand, contention based schemes attempt to resolve collisions when
multiple nodes compete for access to the medium. On the other hand,
scheduling rules prevent collisions by restricting medium access to noninterfering nodes.
In dynamic environments, for example when the wireless nodes are
moving according to some mobility pattern, scheduling the wireless
transmissions is a challenging issue. Mobility affects the link availability at each time instant and the set of topologies the network may evolve
into during the course of time. Since the above information is necessary
to the scheduler, the scheduling decision process gets perplexed.
In our work, we consider scheduling in mobile wireless ad-hoc networks. Our objective is to find a centralized scheduling rule that maximizes the total stable throughput of the network. Our work is inspired
by the work of [3] and [5] and generalizes their results. In [3], the
authors focus on scheduling in a finite node, static multi-hop wireless
ad-hoc network. A centralized, optimal stable throughput stationary
scheduling policy is introduced and its corresponding stability region
is characterized. The authors show that non-stationary scheduling rules

do not achieve a better performance in terms of throughput, and conclude that it suffices to restrict attention to the set of stationary policies.
In [5], Tassiulas addresses the problem of scheduling in a finite node,
time-varying-topology ad-hoc network. However, his model does not
allow multi-hop forwarding of traffic.
In this paper, we generalize the results of [3] and [5] and identify
a family of optimal centralized stationary scheduling policies that all
maximize the network throughput in multi-hop time-varying-topology
networks. Policies within this family are parameterized by the preferential weight assigned to the respective customer classes.
This paper is organized as follows. In section II we introduce our
model. Then, in section III we present the proposed family of centralized scheduling policies. In section IV we give our optimality results.
Then, in section V we confirm our analytical results through a Monte
Carlo simulation. Finally, we conclude the paper in section VI.

II. M ODEL
We consider a slotted time, time-varying-topology wireless, ad-hoc
network, operating under a TDMA based medium access scheme. The
network is comprised of a finite number N of nodes, with mobile capability, each equipped with one transceiver (transmitter/receiver pair) and
an omni-directional antenna. Nodes share a common medium, hence it
is essential to properly coordinate their transmissions to avoid signal
interference and collisions from simultaneously transmitting neighboring nodes. Towards this end, a set of primary constraints is imposed
on the set of simultaneously transmitting neighbors. These constraints
dictate that:
1. A node cannot transmit and receive simultaneously.
2. A node cannot transmit simultaneously to multiple nodes.
3. A node cannot receive simultaneously from multiple nodes.
Nodes exchange datagram packets of constant packet lengths, that
can be transmitted in one time slot, in a unicast fashion. We consider J
distinct customer classes, each intended for a set of exit nodes Vj , j =
1, . . . , J. The set of exit nodes is such that whenever a packet of some
class reaches an exit node for this class, the packet leaves the network.
The different sets Vj are allowed to overlap.
Each node is modeled as a set of J infinite buffer queues, each holding separately the packets corresponding to different customer classes.
We denote by Xij (t) the non-negative integer queue size for class j at
node i at the end of time slot t. In addition, X(t) is a queue length
matrix defined as X(t) = (Xij (t), i = 1, 2, . . . N, j = 1, 2, . . . J) and
for each j {1, 2, . . . J} Xj (t) is a N 1 vector of all queue sizes of
class j at time t, that is, Xj (t) = (Xij (t), i = 1, 2, . . . N ). We denote
by X the space of all queue size vectors.
A link ` is said to exist between nodes i1 and i2 , if node i2 is within
transmission range of node i1 . We model each link as a server. In
particular, a link ` that originates from node s(`) and terminates at node
d(`) is a server that serves a customer from queue s(`) and upon service

completion, sends the served customer (packet) to the destination queue

d(`). All the servers are synchronized to start serving a customer at the
beginning of a time slot. The primary constraints on medium access
make the servers interdependent, in the sense that not all of the servers
can be active at any time slot.
Node mobility causes some already established links to break, as
participating nodes move far away from each other, and results in establishing new connections between nodes that move closer to each other.
Hence, the mobility pattern of the nodes affects the set of possible network topologies, as well as their respective probability distribution.
Finiteness of the number of nodes implies finiteness of the set T of
topologies the network may evolve into, i.e. T = {T1 , . . . , TNT }.
We denote the topology at time slot t by T (t) T . Each topology
Tk is characterized by the set of links L(k) that are present under Tk .
Let L be the set of all links that exist in at least one topology, i.e.
T
L = N
k=1 L(k). A uniform numbering of the links is used across all
topologies, i.e. if link ` {1, . . . , |L|} connects nodes i1 and i2 under
topology k, then in every topology where node i1 is connected to i2 , the
link connecting them is numbered `.
Customers of each class may enter the network at any node, except
for the exit nodes of the corresponding class, and at each time instant.
For each customer class j, the vector of arrivals Aj (t) = (Aij (t) : i =
1, 2, . . . , N, i
/ Vj ) is a component-wise non-negative vector. Its ith
element, Aij (t), represents the number of customers of class j arriving
at queue i, during time slot t. For all queues i : i = 1, . . . , N and all
customer classes j : j = 1, . . . , J, we define aij = E[Aij (t)], that is
assumed to be time-invariant, and we term multi-class arrival vector
the non-negative component-wise matrix a = (aij : i = 1, . . . , N, j =
1, . . . , J).
To capture the dependence among link activations, due to the
medium access constraints, following [3] and [5], we define a valid activation set for topology Tk to be a set of servers (a subset of L(k)),
that comply with the primary constraints and are allowed to be activated
simultaneously. We also define the valid activation vector associated
to an activation set as an |L| element vector with its `th component,
1 ` |L|, set to 1 if the `th server belongs to the activation set
and set to 0 otherwise. If link ` is not present in Tk T or medium
access constraints prevent it from being activated, then the `th element
of all the valid activation vectors for topology Tk must be set to 0. The
constraint set Sk for topology Tk T is the set of all valid activation
vectors associated with network topology Tk . Since the set Sk is determined by the primary constraints, it should be clear that for each k, if
c Sk , all the vectors obtained by setting some of the active links of c
to inactive, belong in Sk as well.
Next, we define the binary random variable E`j (t) {0, 1}, that
indicates whether server ` is active during slot t, serving a customer of
class j. The random variable E(t) = {E`j (t) : ` = 1, . . . , |L|, j =
1, . . . , J} is called a valid multi-class activation vector for topology
T (t), if the corresponding vectors Ej (t) = {E`j (t), ` = 1, . . . , |L|}
are such that:
1. E`j (t) = 0, if Xs(`)j (t 1) = 0.
2. E`j (t) = 0, `
/ T (t).
PJ
j
3.
j=1 E (t) Sk , where T (t) = Tk .
We denote by E the collection of all valid multi-class activation vectors
that correspond to all possible network configurations.
Finally, we define as the state of our system to be the process
{X(t)}
t=0 of the queue sizes at all network nodes and for all classes
of customers at the end of the current time slot. Both the system state

X(t 1) as well as the current topology T (t) are available to the scheduler at the beginning of each time slot t. Based on this information, the
scheduler takes a scheduling decision in a centralized fashion. Accordingly, we define a deterministic, stationary scheduling policy , to be
a map X T E such that, (x, T ) is a valid multi-class activation
vector for the current topology, as this vector is defined previously.

Queue size dynamics

Clearly, the queue size at node i for class j at the end of the next time
slot equals its value at the end of the current time slot, plus the external
arrivals for this class at this node during the next time slot, modified by
internal arrivals or departures for class j that involve node i. Arrivals
and departures are determined by the next multi-class activation vector and by whether or not service completes during the next time slot.
Hence, the queue size process for class j in vector form is:
Xj (t + 1) = Xj (t) + Rj M(t + 1)Ej (t + 1) + Aj (t + 1)

(1)

In Equation 1, Rj is a N |L| matrix that we dub the combined

routing matrix for class j. It has one column for each link ` L and
its elements are:

if d(`) = i, with i
/ Vj
1
j
(2)
ri`
= 1 if s(`) = i

0
otherwise
Matrix M(t) is a diagonal matrix, of dimensions |L| |L|, that captures the link quality. Its `th diagonal element (M (t))` represents the
binary random variable that corresponds to the successful service completion of a customer served by server ` during time slot t. If a customer
completes service and moves from queue s(`) to queue d(`) (or exits
the system if the node where queue d(`) resides belongs to the class
of the exit nodes for the particular customer class), then (M (t))` = 1,
otherwise (M (t))` = 0. The latter case may occur if the `th link is
not present in the current topology at time slot t or when although it is
present, a customer did not receive full service and its service is being
deferred.
Note that rule 1 in the definition of a valid multi-class activation
vector, E(t), guarantees that the elements of X(t), evolving according
to Equation 1, are non-negative at all times.

B Assumptions
Throughout this paper, we make the following assumptions.
Assumption 1 The topology process is a stationary and ergodic process, that is varying in an i.i.d. fashion. In particular pk = P [T (t) =
Tk ], k 1, . . . , NT does not depend on t.
Assumption 2 The process {M (t)}
t=0 of link qualities is an i.i.d process. In addition, for all t, t0 , M (t) and T (t0 ) are independent. Furthermore, E[M(t)|T (t) = Tk ] is independent of t, k {1, . . . , NT }.
We will call this expected value Mk .
Assumption 3 The arrival process {Aij (t)}
t=0 is independent of
the topology process {T (t)}
t=1 and of the service rates process
{M(t)}
t=0 . Also, for a given node i and customer class j the arrival
2
process {Aij (t)}
t=0 is an i.i.d process. Finally, E[Aij (t)] < .

III. A N OPTIMAL THROUGHPUT SCHEDULING RULE

Definition 1 The system is stable if for the queue size process

{X(t)}
t=0 , where X(0) = x Y :

SUPPORTING WEIGHTS

In this section, we introduce a general scheduling policy that treats

differently the customers of different classes. To achieve this class distinction, a weight wj > 0 is assigned to each customer class j. In
addition, we let w = (wj , j = 1, . . . , J) be the vector of of weights of
each class.
Let us define for each link `, the weighted differences between the
source and the destination queues of the link, for all customer classes j,
weighed by the quality of the link and the weight of the corresponding
class. Specifically:
(Dkw (x))`j = wj (xs(`)j xd(`)j )(mk )` ,

(3)

where (mk )` is the `th (diagonal) entry of E[M(t)|T (t) = Tk ].

In addition, let the weight of server `, (Dkw (x))` be the maximum
weighted difference in queue sizes, that is achieved for some customer
class j, i.e.,
(Dkw (x))` = max (Dkw (x))`j .
(4)
j=1,...,J

We also define the weight vector of each link ` that is present under
topology Tk as:

P [y < ] = 1,

y Y

and all states z

i=1 Zi are positive recurrent,

,
if X(t) Y, t > 0
min{t > 0 : X(t)
/ Y }, otherwise

We also define a multi-class arrival vector to be stable if the system is

guaranteed to be stable whenever it operates under this arrival vector.
The modified version of Fosters theorem ([3], [1]) that follows, gives
sufficient conditions for stability of our system.
Theorem 1 (Extended Foster Theorem) [3] Consider a Markov
Chain {X(t)}
t=0 with state space X . Suppose there exists a real valued, bounded from below, function V : X R, a number > 0 and a
finite subset X0 of X such that:
(
E[V (X(t + 1)) V (X(t))|X(t) = x] < , if x
/ X0
E[V (X(t + 1))|X(t) = x] < ,
otherwise
Then, {X(t)}
t=0 is stable in the sense of Definition 1.

w
Dw
k (x) = {(Dk (x))` : ` = 1, . . . , |L|}

Then, the proposed scheduling policy 0w is

j = (jk (x))` , and

1,
wj
( 0 (x, Tk ))` =
(
ck (x))` = 1 and xs(`)j 1,

0,
otherwise,

Optimality of 0w

Let us define the following sets of arrival rates. Initially, we define the

set of arrival rates A:

=
A

k (x))` is given by
where c
T
k (x) = arg max{Dw
c
k (x) c}
cSk

(5)

|L|

J
j
a RN
+ : fk R+ ,

aj =

NT
X

Rj fkj pk ,

k=1

and
(jk (x))` = arg max

j=1,...,J

Dkw (x) `j .

(6)

Ties in obtaining a solution for Equations 5 and 6 are resolved by selecting one of the solutions.
Thus, policies 0w preferentially activate links ` (Equation 5) and
customer classes j (Equations 4 and 6) for which the weighted difference of queue sizes at the source and destination is largest; here the
weights are the products of the probabilities (mk )` of successful transmissions and of the assigned weights, wj . By selecting the links with
the higher differences in queue sizes this family of policies achieves a
load balancing in the network queues: whenever a queue size increases
enough, one of its outgoing links will be activated, thus reducing the
queue sizes.

IV. S TABILITY UNDER RANDOMLY VARYING TOPOLOGIES

A network is stable if the queue lengths reach a steady state and
do not tend to increase without bound. Under Assumptions 1 - 3, the
process of the queue sizes described by Equation 1, is a Markov Chain.
However, this Markov Chain is not guaranteed to be irreducible.
Let Zi , i = 1, 2, 3, . . . be the sets of its recurrent communicating
classes and Y be the set of all transient states. Following [3] we define
stability of the network as follows:

J
X

fkj co(Mk Sk ) ,

j=1

Next, we define a slightly smaller set A? as follows:

A? =

|L|

J
j
j
a RN
+ : fk R+ , a =

NT
X

Rj fkj pk

k=1

J
X

fkj co(Mk Sk ), for some > 1 .

j=1

Let C be the set of arrival rates for which the stationary scheduling
policy stabilizes the system. In addition, define C to be the set of
arrival rates for which there exists some stationary policy that stabilizes
the system,i.e
C = C
Then, we have proved the following theorem.
Theorem 2 The following properties hold:
is a convex polytope.
1. A
C and C0 are related
2. For every w with wj > 0, j, A? , A,
through the following set inequality:
A? Cw C A
= C w
relint(A)
0
0

(7)

The proof of Theorem

Theorem 1 with function V : X
P 2 invokes
P
2
R, such that V (x) = Jj=1 N
w
i=1 j {xij } . Theorem 2 implies that
and differs from
the set A? contains the relative interior of polytope A
it at most by points on its faces. In addition, it implies that C and C0w

are almost identical and almost identical to the convex polytope A.

only by missing certain
Specifically, C and C0w may differ from A
viewed as a subset of the affine space A it
points from its faces (with A
generates). In particular, C and C0w have equal measure in A. In that
sense, 0w is (near) optimal in the sense of maximum throughput.

90

2
180

45

(t)

V. S IMULATION RESULTS
In this section, we will consider a simple example of a time-varying
topology network and characterize its stability region under the proposed scheduling scheme. We will observe that under a policy 0w although the total throughput under any of the individual topologies the
network takes is zero, the throughput of the time-varying network that
evolves through each one of them with non-zero probability is strictly
positive. Hence, mobility increases the maximum stable throughput that
the network can sustain under 0w .
We assume that the servers are perfect, implying that M (t) is the
identity matrix for all t. The policy that is simulated is 01 , in other
words for each customer class wj = 1, j = 1, . . . , J. Our graphs are
obtained through a Monte Carlo simulation. The arrivals are Bernoulli.
We start by selecting an arrival rate vector randomly according to the
uniform distribution, within the interval (0, 1]. We then run the simulation for 10, 000 time slots. We decide instability on a threshold based
criterion, i.e. if any of the network queues exceeds a threshold under
some multi-class arrival vector, this rate vector is said to be unstable.
If the rate vector is deemed unstable, we decrease some element of this
vector by a step = 0.05 and start the simulation again. However, if it
is a stable arrival rate vector, then we mark it by a dot (Figure 5) and
a new arrival rate is again selected randomly according to the uniform
distribution in (0, 1].

4 nodes on a ring

Consider a network comprised, of 4 nodes that reside on a ring of radius

r. Assume two customer classes: customer class 0, of rate a10 , arriving
at node 1, with exit node 3 and customer class 1, of rate a41 , arriving at
node 4, with exit node 2. Further, assume that at each time instant the
positions of all the nodes are determined by the position of any one of
them. Specifically, let a node be located at angle (t) [0, /2], t
0. The rest of the nodes will be placed at positions (t), +(t) and
2 (t). Hence, each quadrant is allocated to exactly one node. Let
the nodes be numbered from 1 to 4, as shown in Figure 1. Consider also
that the power at the nodes is such, that a pair of nodes communicates
if their distance is at most R. Assume that the radius of the ring is large
enough to guarantee the existence of some region where there exists no
connectivity among the nodes, in other words r > R2 .
If (t) is small enough, then the network nodes that may communicate with each other will be the pair of nodes 1 and 2, as well as nodes
3 and 4. This is illustrated in Figure 2. This would put a constraint
on the distance between these nodes to be less than the range of com
R
).
munication, namely 2rsin((t)) R, i.e. (t) 0 = sin1 ( 2r
Hence, as soon as (t) increases and becomes larger 0 , the distance
between communicating nodes 1 and 2 and 3 and 4 becomes larger than
R, and hence they may not communicate any more. More specifically,
the network will be completely disconnected for values of (t), such
that 0 < (t) < /2 0 . This is illustrated in Figure 3. Further
increase in (t), i.e. /2 0 < (t) < /2 results in a different set

270

Figure 1: A network of 4 nodes, residing on a ring of radius r.

V1

10

41

Figure 2: Topology T0 is present. The nodes that are able to

communicate with each other are nodes 1 and 2 and nodes 3 and
4.

of communicating nodes, namely 1 communicates with 4 and 2 with 3,

as shown in Figure 4. So we have the following set of possible network
topologies:
0 (t) 0 : Topology T0 is present (Figure 2).
0 < (t) < 2 0 : Topology T2 is present (Figure 3).
0 (t) /2: Topology T1 is present.(Figure 4).
By considering a uniform distribution on (t), we observe that
0
topologies T0 and T1 occur each with equal probability p = 4
and
2
T2 occurs with probability 1 2p. In each of the individual topologies and since sources are completely isolated from their intended exit
nodes the maximum throughput of the network is 0, for both customer
classes. However, by considering a network that switches among the
three topologies, taking each one of them with a non zero probability,
we observe that a strictly non-zero stability region is achieved. In particular, consider the example when p = 0.25. Then, the maximum

12p

10

41

Figure 3: Topology T2 is present. All network nodes are disconnected from each other.

the fact that relaying through intermediate nodes can be used to create
paths from source nodes to their corresponding exit nodes.

R EFERENCES

10

[1] Pierre Bremaud. Markov Chains: Gibbs Fields, Monte Carlo Simulation
and Queues. Springer, 1999.

41

Figure 4: Topology T1 is present. Nodes 2 and 3 communicate

and nodes 1 and 4 communicate.
Stability region for a network of 4 nodes arranged on a ring
1
Stable operating points from simulation
Linear least squares fit
Stable throughput region

stable arrival rates of customer class 1 (pkts/slot)

0.9

[3] L. Tassiulas and A. Ephremides. Stability properties of constrained queueing systems and scheduling policies for maximum throughput in multihop
radio networks. IEEE Transactions on Automatic Control, 37(12), December 1992.
[4] Leandros Tassiulas. Dynamic link Activation Scheduling in Multihop Radio
Networks with Fixed or Changing Connectivity. PhD thesis, University of
Maryland-College Park, 1991.

0.8

[5] Leandros Tassiulas. Scheduling and performance limits of networks with

constantly changing topology. IEEE Transactions on Information Theory,
43(3), May 1997.

0.7

0.6

0.5

0.4

0.3

0.2

0.1

[2] Anna Pantelidou. Scheduling transmissions in wireless ad-hoc networks

with time-varying topologies. Master of science, Department of Electrical
and Computer Engineering, University of Maryland, College Park, December 2004.

0.1

0.2

0.3
0.4
0.5
0.6
0.7
stable arrival rates of customer class 0 (pkts/slot)

0.8

0.9

Figure 5: Stability region of the network consisting of 4 nodes

on a ring, when the occurrence probability of topologies T0 and
T1 is p = 1/4.

throughput scheduling policy 01 as simulated by our simulation tool,

gives stable points that are marked with blue dots (Figure 5). Clearly
every point that lies in the first quadrant and is bounded from above
these points is stable as well. In Figure 5, we also drew the linear
least square fit for the set of stable points that is depicted by the line
a10 + a41 = 0.5. The shaded region below the line a10 + a41 = 0.5 is
the total stability region of our system. Our simulation results coincide
what we would guess intuitively about the stability region of this simple time-varying topology network. For example, consider the customer
class 0. Although the links from node 1 to 2 and from node 2 to 3 are
available 0.25 fraction of time each, we can make the arrival rate a10 to
be stable for values strictly larger than 0.25 by letting node 1 forward
its traffic through both nodes 2 and 4.

VI. C ONCLUSIONS
In this paper, we introduced a family of centralized stationary
scheduling policies applied on time-varying-topology wireless ad-hoc
network. We characterized the common set of stable arrival rates that all
of these policies achieve. In addition, we proved optimality of all policies in this family in terms of maximizing the total stable throughput of
the network. We did that by showing that the set of rates that are stable
under any member of this family is essentially identical to the union of
the sets of stable arrival rates achieved by all stationary policies. Finally, we confirmed through simulations that mobility at the nodes may
indeed increase the total stable throughput of the network. This is due to