CONFIGURABLE TASK MAPPING FOR MULTIPLE OBJECTIVES IN MACRO-PROGRAMMING OF WIRELESS SENSOR NETWORKS

INTERNATIONAL JOURNAL OF ADVANCED SMART SENSOR NETWORK SYSTEMS ( IJASSN ), VOL 1, NO.1, APRIL 2011
1
CONFIGURABLE TASK MAPPING FOR MULTIPLE
OBJECTIVES IN MACRO-PROGRAMMING OF
WIRELESS SENSOR NETWORKS
Himanshu Singh1
and R.S. Singh2
1,2
Department of Computer Engineering, Institute of Technology, Banaras Hindu
University, Varanasi, India.
1himanshu.singh.cse07@itbhu.ac.in, 2ravi.cse@itbhu.ac.in
ABSTRACT
Macro-programming is the new generation advanced method of using Wireless Sensor Network (WSNs),
where application developers can extract data from sensor nodes through a high level abstraction of the
system. Instead of developing the entire application, task graph representation of the WSN model presents
simplified approach of data collection. However, mapping of tasks onto sensor nodes highlights several
problems in energy consumption and routing delay. In this paper, we present an efficient hybrid
approach of task mapping for WSN – Hybrid Genetic Algorithm, considering multiple objectives of
optimization – energy consumption, routing delay and soft real time requirement. We also present a
method to configure the algorithm as per user's need by changing the heuristics used for optimization.
The trade-off analysis between energy consumption and delivery delay was performed and simulation
results are presented. The algorithm is applicable during macro-programming enabling developers to
choose a better mapping according to their application requirements.
KEYWORDS
Wireless Sensor Networks, Macro-programming, Genetic Algorithm, Task Scheduling, Energy
Conservation
1. INTRODUCTION
Wireless Sensor Network is becoming a popular research field in distributed systems due to its
wide applications. Consider the network requirements of a particular farm scenario. A farmer,
having different fields at different locations, may wish to collect normal data about the field
conditions for example – temperature, water level in the soil and humidity continuously at
regular intervals; and would also like to be informed about any event happening on the field like
sudden flood and fire. A wired information system would be too costly for this case. The most
feasible solution for such a system would be to use a wireless sensor network across all the
fields where nodes have various sensors (also known as motes) attached to it for taking the
measurements and a radio for wireless communication. To maximize the network lifetime,
energy consumption by batteries should be minimum since it is not feasible to replace the
batteries of nodes once deployed in farm. Also, the data delivery time should be very low,
especially for fire and flood notifications.
Now, the farmer may not be educated enough to program the sensor nodes in WSN specific
operating systems like TinyOS [1], Contiki [2], MantisOS [3] and to optimise the above
mentioned parameters. Macro-programming [4] constructs allow the farmer to specify his
requirements in the form of a task graph and a network graph. The tasks may be different types
of sensing tasks, filtering tasks, routing tasks etc which may be specific to the requirements of a
particular field. The network graph consists of network of sensor nodes on which the tasks have
to be executed. Some of the macro-programming architectures like KairOS[5], Regiment [6]
and COSMOS [7], which provide such high level abstractions to define the system model, face

2
the problem of efficient mapping of one or more tasks on some sensor nodes. In this paper, we
study this task mapping problem using the Genetic Algorithm [8]. The heuristic used is based
on the optimization over multiple parameters –total energy consumption, efficient routing, and
soft real time delivery requirements. In section 2, we present previous research in this area and
their difference with our work. In section 3, we formulate the problem and then present the
Hybrid Genetic Algorithm (HGA) in section 4. Simulation results of trade-off analysis have
been shown in section 5 and we conclude in section 6 illustrating how different farmers can
control optimization of one parameter (like energy) on the cost of others (like delivery time).
2. RELATED WORKS
Task mapping is a widely studied problem in parallel and distributed systems. But in wireless
sensor networks, it gained attention of researchers due to energy constraints and delivery time
requirements. Most of the researches [9], focus on reducing the energy consumption by efficient
MAC protocols which introduce sleep-awake scheduling of nodes. Unlike our work, they do not
consider the possible optimization in routing and efficient task allocation. The work in [10]
addresses the problem of task placement on sensor nodes, but they consider single hop networks
only. The work in [11] also present the task mapping problem, but they focus only on the
objective of minimizing the total energy consumption. With macro-programming becoming a
popular field, work in [12] adopted a greedy method of task mapping but it maps the tasks only
on the tree topology and not general network graphs. Some macro-programming languages like
KairOS present the concept of dividing the flow of program into 'Control Flow Graphs' (CFG)
and mapping CFGs onto sensor nodes. KairOS minimizes total number of edges that crosses
CFG. It does not clearly offer any solution to energy constraints. [13] offers a solution based on
greedy method particularly for macro-programming, but they do not offer any method to
configure the optimization according to needs of application. COSMOS[7] and Srijan [14]
which use mOS and mPL languages, program the WSNs using the concept of task graph. These
applications do not provide the support of configuring the task mapping in application; however
these act as perfect platforms wherein the HGA algorithm (presented in section 4) can be used
to map the task graphs on motes with optimizations over multiple-objectives and allowing users
to configure these optimizations by changing the heuristics used.
3. PROBLEM IDENTIFICATION
Tasks which need to be mapped onto sensor nodes, can be sensing tasks, routing tasks,
activation tasks etc. Since these tasks need to execute in a particular order (Sensing, then
routing, then activation) the tasks can be modelled as a directed acyclic graph. The sensor node
which form a wireless network, are modelled as undirected network graph.
3.1. Objectives
Objective of this work is to obtain a mapping of every task into some sensor nodes, then
ascertaining the routing path between the predecessor and successor and finally ascertaining
start up time and scheduling order of every task.
Figure 1: Task Graph needs to be mapped on Network Graph

3
Multi-objective optimization: In a real time environment, it becomes of a greater necessity to
execute certain tasks with higher priority within given deadline and ensuring that minimum
energy is consumed in completing all the tasks. Such requirements need simultaneous
optimization of multiple objectives. Solutions to such problems are usually computed by
combining multiple objectives into a single criterion to be optimized. The combining of
objectives has the advantage of producing a single compromised solution, and normally does
not require any human intervention. However, the problem is that if the optimal solution cannot
be accepted because we chose an improper setting of the coefficients of the combining function,
user intervention may be required. Such interaction should be easy and high-level. Therefore,
we consider finding an optimal schedule as a problem of multi-objective optimization with non-
measurable objectives. Genetic Algorithms are capable of multi-objective optimization. The
algorithm explores a set of solutions that can only be improved in one way by being degraded in
another instead of collapsing entirely. Therefore, we use a multi-objective GA which
independently considers multiple objectives, and finds a set of solutions that satisfy all of the
following objectives and simultaneously allows users to increase/decrease optimization of any
parameter.
1. Minimize total schedule time cost
2. Minimize total schedule energy cost
3. Minimize tardiness value to achieve soft real time deadline
3.2. Directed Acyclic Task Graph
Let the task graph be TG= (V, E), where V be the set of vertices and E be the connecting edges.
A vertex Vi corresponds to a task Ti. Each Ti is identified by (Energyi , Timei ) where Energyi
denote the energy and Timei is the time required for execution of Ti.. Eij connecting <Vi, Vj>
denotes that the Vj could not start until and only until the task Vi has been executed. The weight
on the directed edge Ci, j between Vi and Vj denotes the amount of data transmitting from task Ti
to Tj. Figure 1 illustrates a DAG task Graph T.
Figure 2: Task Graph
3.3. Network Graph
The network graph NG= (P, L) is used to denote the wireless sensor network. P is the set of
sensor nodes which process the some tasks. L is the set of communicating undirected edges. The
weight of edge dist_nodesi j, comm_time_costi j, wi j respectively denotes the distance, the
communicating delay, the energy consumption for transmitting one unit data by vertex Pi and
receiving one unit data by vertex Pj when data is transmitted between vertex Pi and Pj.

4
Figure 3 (a) Network Graph denoting distance (b) Network Graph denoting energy
In Figure 3(a), the data on the edge denote the physical distance between two nodes. In Figure
3(b), the data on the edge denote the energy consumptions for one unit transmitting of data.
We use the energy model presented in [15]. Energy is expended to serve the purpose of: (1)
digital electronics, E_ELEC, (actuation, sensing, signal emission/reception), and (2)
communication, EAMP. EAMP varies according to the distance dist_nodes between a sender
nodes and a receiver. To transmit k bits for a distance dist_nodes, the radio expends k ×
(E_ELEC + EAMP × dist_noden
) J, where n = 2 for dist_nodes < Do, and n = 4 for dist_nodes
>= Do, and Do is a constant threshold distance which depends on the environment. To receive k
bits at the receiver, the radio expends k × E_ELEC J.
4. HYBRID GENETIC ALGORITHM
4.1. Background and Definitions:
Let initial energy of each node is e0. As the system operates, the energy remaining at node k be
ek. The Execution Cost Matrix E is a single dimensional matrix that represents energy cost of
execution of task i. Transmission Cost (Etrans) is calculated as the energy required/spent to
transmit data from every single node to every other node in the network.
Transmission energy cost, when Vi transmits data to Vj, is the sum of the energy consumption
for transmitting unit data (Wm n = E_ELEC + EAMP × dist_noden
) on every edge in path of Vi
and V j taken product with data Ckl.
The whole energy consumption “Energy” of accomplishing task is the sum of the energy spent
in completing task ti on nodej and transmitting data. That is:
To formulize the objective of meeting the deadline of each task, let us assume that we have a
task, say ti, which has an execution time (ei) and a deadline (di).
We consider two types of times: earliest and latest start times of each task. The earliest
start time of task ti (estStT[ti]) is the length of the longest path from the entry task to ti.
The latest start time of task ti (lstStT[ti]) is defined as the latest time at which ti can start,
such that ti and all successors of ti have a chance of meeting their deadlines.

5
For achieving soft real time deadlines, our approach considers the communication time between
two tasks and time taken to execute those tasks in calculating the latest start times of tasks.
Tardiness of a task is defined as the difference between actual start time in schedule and latest
start time of that task. The cost of a schedule is defined in terms of sum of maximum tardiness
among all tasks. The objective is to find a schedule with the maximum tardiness minimized.
4.2. Algorithm
4.2.1. Algorithm 1 - Hybrid Genetic Algorithm
Input: Network topology, DAG task graph, initial population size: init_population, evolutional
generation: T.
Output: Task mapping scheme, scheduling scheme and routing scheme.
Step1: i=1; // evolutional generation count
Step2: generate initial GA_ population (Pop) using create_initial_Chromosome
Step3: Calculate schedule_time(), schedule_energy(), tardiness() of Population
Step4: Repeat
Step5: calculate the fitness value f(Pop[i]) of the individuals in population using
calculat e_fitness. // computing individual fitness
Step6: [child1, child2] = Roulette_Selection(Pop[i]);
// Select the individuals having optimal fitness value using Roulette criteria[16]
Step7: Pop[i+1] ← OptimalSelect(Pop[i]); // selection operation
Step8: GTi ← Hybrid_operation(child1, child2); //hybrid operation
Step9: GSi+1 ← Evolove(GTi ) //mutation evolution operation
Step10: i=i+1; Calculate schedule_time(), schedule_energy(), tardiness() of newly
generated population
Step11: until (i>Threshold) or (termination criterion is met).
Step12: Ouput Best individual found so far;
4.2.2. Algorithm 2 – create_Initial_Chromosome
Supposing the number of DAG tasks which are scheduled to m sensor is n. The allowable
energy deviation is E.
Step1: n tasks are mapped to sensors.
Step2: mapping a routing path for every edge of DAG
Step3: Get(i，v1，v2)；//the start point and endpoint of the edge i is respectively mapped to
nodes v1 and v2 in sensor Network.
If v1=v2 then there is no use for routing
Else
Find (v1,v2,k) // Use Floyd-warshall algorithm to find shortest path
Step4: Select (v1,v2,k); // select the least energy cost path from the k paths.
End
4.2.3. Algorithm 3 – Calculate_fitness(Pop[i])
Step1: TMax = Maximum time of execution of tasks in Pop;
EMax = Maximum energy of execution of tasks in Pop;
TDMax = Maximum tardiness of all tasks.
Step2: Fitness of ith
Chromosome, pop[i].fitness = W1*fitness_time/Tmax +
W2*fitness_energy/Emax + W3*fitness_tardiness/TDmax; where fitness_time = Tmax -
pop[i].schedule_time and W1, W2, W3 are user defined weights of different objectives.

6
4.2.4. Algorithm 4 – Hybrid_operation(child1, child2)
Step1: //Crosses over two individuals pop[child1] and pop[child2]
Step2: if (random(0,20) > 15) then num_bits = random(M/2,M) ; //M is no. of tasks
else num_bits = random(0,M/2) ;
Step3: for i = 0 to num_bits //Mutation between child1 and child2
pos = random(0,M) ;
temp = pop[child1].schedule[pos] ;
pop[child1].schedule[pos]=pop[child2].schedule[pos] ;
pop[child2].schedule[pos] = temp ;
End
4.2.5. Algorithm 5 – Evolve((offspring)
M is the number of tasks
Step 1: Initialize num_bits = 0.
Step 2: if (random(0,20) > 15)
num_bits = random(M/2,M)
else
num_bits = random(0,M/2)
Step 3: for ( i=1 to num_bits)
Assign a random node to a random node;
5. PERFORMANCE AND ANALYSIS
5.1. Simulation scene and parameters
The HGA algorithm was simulated by varying the number of tasks and number of nodes. The
energy cost of performing tasks is taken randomly between 1000J and 5000J, the data to be
transferred is a random number between 5 to 50 bits, and the time delay for communicating
between two nodes is a random number between 1 and 5. Distance between two nodes is a
random number from 10 to 100 units. The transmission energy is calculated as described in
section 3.2. When the energy of a node becomes less than 0 J, the node is believed as dead and
is excluded for further scheduling in particular population. The weights of optimization W1
(weight of energy optimization), W2 (weight of total time of execution of tasks in the schedule),
W3 (weight of total tardiness of all tasks) are varied and based on simulation using these
parameters; we now discuss the results in next subsection.
5.2. Results
5.2.1. Comparison of Total Schedule Time
As observed in Figure 4 and 5, when energy minimization is priority in figure 5, the schedule
time using Hybrid objective is larger than that in figure 4 where time minimization is priority.
To illustrate, Figure 4 shows the total time of execution of all tasks when W2 (weight of time
requirement) is larger than W1 (weight of energy optimization), whereas Figure 5 shows total
time of execution when W1 is larger i.e. energy optimization is priority.

7
Figure 4. Best solution with minimum time only Vs Best solution with minimum energy only
Vs Best Solution using Hybrid objectives with time as priority
Vs Best Solution with Hybrid objectives with energy as priority
5.2.2. Comparison of Total Schedule Energy
Figure 6 and 7 show total energy consumption of solution. The energy consumption is larger
when time minimization is priority for the user, shown in Figure 6, than when energy
minimization is priority, shown in Figure 7.
Vs Best Solution using Hybrid objectives with time as priority

8
Vs Best Solution with Hybrid objectives with energy as priority
5.2.1. Comparison of Total Tardiness
When objective of achieving soft real time deadlines are included, the tardiness value of best
schedule decreases as shown in Figure 8.
Figure 8. Best Solution using Hybrid Algorithm without considering tardiness value Vs Best
Solution w.r.t energy using Hybrid Algorithm considering tardiness value
6. CONCLUSIONS
For scheduling tasks with limited energy in WSN, we implemented Hybrid Genetic Algorithm
that will find a mapping considering many objectives. Based on heuristic used – fitness value,
users can configure the mapping for a particular objective, just by changing the weights W1,
W2, W3 etc [Section 4.2.1], thus changing the underlying routing and mapping scheme.
Although Genetic Algorithm requires high computing, the algorithm will be useful particularly
in macro-programming with on-the-fly reprogrammable sensor nodes where only one time
mapping is required during macro-programming.
REFERENCES
[1] P. Levis, S. Madden, J. Polastre, R. Szewczyk, K. Whitehouse, A. Woo, D. Gay, J. Hill, M.
Welsh, E. Brewer, and D. Culler, “TinyOS: An operating system for wireless sensor networks,”
in Ambient Intelligence. New York, NY: Springer-Verlag
[2] Dunkels, A., Gronvall, B., and Voigt, T. 2004. “Contiki - A Lightweight and Flexible Operating
System for Tiny Networked Sensors.” In Proceedings of the 29th Annual IEEE international
Conference on Local Computer Networks (November 16 - 18, 2004). LCN. IEEE Computer
Society, Washington, DC.

9
[3] Bhatti, S., Carlson, J., Shucker, B., Gruenwald, C., Torgerson, A., and Han, R. 2005. “MANTIS
OS: an embedded multithreaded operating system for wireless micro sensor platforms.” Mob.
Netw. Appl. 10, 4 (Aug. 2005).
[4] A. Pathak and V. K. Prasanna, “Energy-efficient task mapping for data-driven sensor network
macro-programming,” in International Conference on Distributed Computing in Sensor Systems
(DCOSS), June 2008.
[5] R. Gummadi, O. Gnawali, and R. Govindan, “Macro-programming wireless sensor networks
using kairos,” in International Conference on Distributed Computing in Sensor Systems
(DCOSS). Springer, 2005, pp. 126–140.
[6] R. Newton, G. Morrisett, and M. Welsh, “The regiment macroprogramming system,” in IPSN
’07: Proceedings of the 6th international conference on Information processing in sensor
networks. New York, NY, USA: ACM, 2007, pp. 489–498.
[7] A. Awan, S. Jagannathan, and A. Grama,“Macroprogramming hetero-geneous sensor networks
using cosmos,” in EuroSys ’07: Proceedings of the 2nd ACM SIGOPS/EuroSys European
Conference on Computer Systems 2007. New York, NY, USA: ACM, 2007, pp. 159–172.
[8] Haupt, R.L., Haupt, S.E., “Practical genetic algorithms”, John willy & Sons, 2004.
[9] Li, Y., Ye, W. and Heidemann, J. “Energy and latency control in low duty cycle MAC
protocols.” In Proceedings of the IEEE Wireless Communications and Networking Conference,
New Orleans, LA, USA, March 2005. SMAC and TMAC
[10] Y. Yu and V. K. Prasanna, “Energy-balanced task allocation for collaborative processing in
wireless sensor networks,” MONET, vol. 10, no. 1-2, pp. 115–131, 2005
[11] Y. Tian, E. Ekici, and F. Ozguner, “Energy-constrained task mapping and scheduling in
wireless sensor networks,” in IEEE International Conference on Mobile Adhoc and Sensor
Systems Conference RPMSN Workshop, 2005.
[12] Z. Abrams and J. Liu, “Greedy is good: On service tree placement for in-network stream
processing,” in ICDCS ’06: Proceedings of the 26th IEEE International Conference on
Distributed Computing Systems. Washington, DC, USA: IEEE Computer Society, 2006, p. 72.
[13] A. Pathak and V. K. Prasanna, “Energy-efficient task mapping for datadriven sensor network
macroprogramming,” in International Conference on Distributed Computing in Sensor Systems
(DCOSS), June 2008.
[14] A. Pathak and Mahanth K. Gowda, “Srijan: a graphical toolkit for sensor network
macroprogramming”, In proceedings of the 7th joint meeting of the European software
engineering conference and the ACM SIGSOFT symposium on The foundations of software
engineering, ACM , New York, USA, 2009.
[15] T. Rappaport, “Wireless Communications: Principles & Practice”. Englewood Cliffs, NJ:
Prentice-Hall, 1996.
[16] Fitness proportionate selection, Wikipedia, Retrieved on 2nd
February, 2011.
http://en.wikipedia.org/wiki/Fitness_proportionate_selection
Authors
Himanshu Singh is an undergraduate student at
Institute of Technology, B.H.U, India.
His research interests include Computer
Networks and Wireless Sensor Networks.
R. S. Singh is Assistant Professor at Institute of
Technology, B.H.U, India His research
interests include Graph Algorithms and
High Performance Computing.

CONFIGURABLE TASK MAPPING FOR MULTIPLE OBJECTIVES IN MACRO-PROGRAMMING OF WIRELESS SENSOR NETWORKS

More Related Content

What's hot (17)

Similar to CONFIGURABLE TASK MAPPING FOR MULTIPLE OBJECTIVES IN MACRO-PROGRAMMING OF WIRELESS SENSOR NETWORKS (20)

Recently uploaded (20)

CONFIGURABLE TASK MAPPING FOR MULTIPLE OBJECTIVES IN MACRO-PROGRAMMING OF WIRELESS SENSOR NETWORKS