HYBRID GENETIC ALGORITHM FOR BI-CRITERIA MULTIPROCESSOR TASK SCHEDULING WITH COMMUNICATION DELAY

Advanced Computational Intelligence: An International Journal (ACII), Vol.3, No.4, October 2016
DOI:10.5121/acii.2016.3403 19
HYBRID GENETIC ALGORITHM FOR BI-CRITERIA
MULTIPROCESSOR TASK SCHEDULING WITH
COMMUNICATION DELAY
Sunita Dhingra1
*, Ashwani K. Dhingra2
,SatinderBal Gupta3
and Ranjit Biswas4
1
Department of Computer science & Engineering, 2
Department of Mechanical
Engineering,
University Institute of Engineering & Technology,
MaharshiDayanand University Rohtak-124001 Haryana, India
3
Department of Computer Science, Vaish College of Engineering Rohtak-124001
Haryana, India
4
Department of Computer Science & Engineering, JamiaHamdard University New
Delhi-110062 , India
ABSTRACT
Present work considers the minimization of the bi-criteria function including weighted sum of makespan
and total completion time for a Multiprocessor task scheduling problem.Genetic algorithm is the most
appealing choice for the different NP hard problems including multiprocessor task scheduling.
Performance of genetic algorithm depends on the quality of initial solution as good initial solution provides
the better results. Different list scheduling heuristics based hybrid genetic algorithms (HGAs) have been
proposed and developedfor the problem. Computational analysis with the help of defined performance
index has been conducted on the standard task scheduling problems for evaluating the performance of the
proposed HGAs. The analysis shows that the ETF-GA is quite efficient and best among the other heuristic
based hybrid genetic algorithms in terms of solution quality especially for large and complex problems.
Keywords:
Multiprocessor task scheduling, heuristics, Hybrid Genetic algorithm, makespan, Total Completion time.
1.INTRODUCTION
Parallel processing is the most promising approach for meeting the increased computational
requirements thathas introduced a number of problems including multiprocessor task scheduling.
Usually a large program is divided into smaller tasks havingsome dependencies representing the
precedence constraints such that a task cannot be started until all its predecessors have finished.
The goal of a task scheduling algorithm is to schedule all the tasks on the given number of
available processors without violating precedence constraints so as to minimize the different
performance measures in order to maximize the throughput and utilization of the system. The
multiprocessor task scheduling problem with precedence constraints belongs to the class of NP
hard problems and its importance led to several comparative studies. The conventional research
on scheduling considers heuristic based algorithms which are divided into three categories: list
scheduling, clustering based and duplication based. Among heuristic algorithms, List scheduling
algorithms are commonly used and work by assigning priority to each task according to attributes
such as t-level (top level), b-level (bottom level), static level (sl) and ALAP (As Late As Possible)

20
start time[1-5,10].The task with highest priority is assigned to the processor which gives the
earliest start time. Task selection can be random or based on some rules for the task having the
same priority. The heuristic algorithms follow the procedure that narrow the search to a very
small portion of the solution space. Heuristic-based approaches have greedy nature due to which
they are are not likely to produce reliable results for a wide rangeof problems, particularly when
the density of the scheduling problem increases.
Therefore, researchers go for metaheuristics such as GeneticAlgorithm,Simulated annealing,
Tabusearch, Particle swarm optimization etc.for the sake of better quality schedules.Most of the
metaheuristicsespecially Genetic Algorithm(GA) outperformed traditional heuristic based
scheduling algorithms at the cost of extra time andcomputing effort[6].It is therefore
hybridization of metaheuristicswith heuristics is the next choice for improving the solution
quality.
In the present work, an attempt has been made for hybridization of different list scheduling
heuristics with genetic algorithm for the multiprocessor task scheduling problem with precedence
constraints on homogeneous processors for minimizing the considered bi-criteria objective
function. The remaining part of the paper is organized as follows: section 2 is based on literature
review in the fieldof hybrid algorithms related to task scheduling. Problem formulation along
with assumptions is presented in section 3. Section 4 describes the different hybrid Genetic
Algorithm (HGAs). Section 5 gives analysis of experimental results with discussions followed by
the conclusionin section 6.
2. RELATED WORK
As the complexity of the scheduling problem increases the heuristic algorithms fails to provide
the reliable results.Metaheuristic based scheduling algorithms obtain schedules of better quality
but at the expense of more computing efforts due to limited exploration ability.Houet al.
[6]proposed the first and most important work that has used GA for multiprocessor tasks
scheduling which uses theheight of tasks in input DAG.Though the algorithm is very simple in
terms of computational complexity, but it cannot guarantee that the search space is global due to
which someviable schedules are not accessible [11]. Houetal.[6] andDhodhi& Ahmad [7]
developed a new technique in combination with the ISH [2] and DSH [2] based list scheduling
heuristicswith GAs and called as problem-space genetic algorithms(PSGAs).The analysis of
results showed that the GA when combined with heuristics could work efficiently and schedule
the taskson several processors. In literature, different methods have also beenattempted to
combine the heuristic and genetic approaches for the solution of theproblem [7, 8 23].
Sivanandam et al. [13]proposed a hybrid algorithm in support of particle swarm
optimization/simulated annealing (PSO/SA) for static allocation of tasks in a heterogeneous
distributed computing system with the objective of minimizing the cost. Implementation has been
carried out on different PSO algorithm with Simulated Annealing. Different experiments have
been performed on the benchmark problems and shows that the proposed hybrid method was
effective and efficient in finding near optimal solutions.Yooet al. [14] proposed a multi-objective
hybrid genetic algorithm (MOHGA) for real-time tasks on heterogeneous multiprocessor
environment with the purpose of minimizing the total tardiness and completion time
simultaneously. The adaptive weight approach has been used for multiple objectives. The
convergence of GA is enhanced by introducing the probability of SA as the measure for taking

21
new trial solution. The effectiveness of proposed algorithm has been checked by various
experiments andit was concluded that the MOHGAprovides improved resultsas compared to
other algorithms without communication cost.Dahal et al. [15]proposed a hybrid algorithm for
dynamic scheduling. Thewell known heuristics such as ‘Earliest Deadline First (EDF)’ and
‘Shortest Computation Time First (SCTF)’ has been hybrid with the Genetic Algorithm (GA). It
was concluded that hybridization of GA with SCTF provides better performance as compared to
the EDF based hybrid GA.Azghadi et al. [16] developed an immune genetic approach for
multiprocessor task scheduling problem and proves to be more effective.Jouglet et al. [17]
proposed a memetic algorithm (MA) for the hybrid flow shop scheduling with multiprocessor
tasks. They hybridized theGenetic Algorithm (GA) and Constraint Programming (CP) for a
Memetic Algorithm (MA) and concluded the superiority of proposed MA.Hwang et al. [18]
proposed priority-based GA which uses a new encoding mechanism with a multi-functional
chromosome that uses the priority representation— so-called priority-based multi-chromosome
(PMC). They addressed the problem of multiprocessor task scheduling with communication cost.
From results, it is concluded that proposed priority-based GA has effective performance in
various parallel environments for scheduling methods.
Kim et al. [19]considered heuristic method for a deterministic scheduling problem where multiple
jobs with s-precedence relations are processed on multiple identical parallel machines for
minimization of the total completion time.Goh et al. [21]considered a heterogeneous
multiprocessor scheduling problem with precedence constraint and proposed a hybrid
evolutionary algorithm (HEA) formakespan minimization. The method considered the partial list
scheduling and duplication scheduling heuristic for exploiting the intrinsic structure of the
solution and specialized genetic operators for promoting the exploration of the search space.
Experiments were carried on a set of benchmark problems and it was concluded that the
proposed HEAprovide better results. Wen et al. [22] incorporated GA with both Variable
Neighborhood Search (VNS) and a heuristic extracted from traditional list scheduling algorithms
for the minimization of makespan in the heterogeneous multiprocessor scheduling problem
resulting into a heuristic based hybrid genetic variable neighborhood search algorithm. The
performance of proposed approach was compared with four related algorithms, HEFT, AIS, VNS
and IGA on standard benchmarks problems and it was concluded that proposed algorithm
constantly outperforms the other four algorithms in terms of schedule quality.
Mohamed et al.[23]proposed the Modified List Scheduling Heuristic (MLSH) along with the
hybridization with theGenetic Algorithm for multiprocessor task scheduling system and
concluded the superiority when compared to others.Roy et al. [24]considered task scheduling in
multiprocessor systems and proposed a heuristic based Genetic Algorithm by choosing the
eligible processor on educated guess. Variation of HLFET algorithm with genetic algorithm was
proposed and experiments were performed on Standard Task Graphs (STG).It was concluded that
the algorithm has better average makespan than Elitism stepping method in lesser number of
evaluations.
Hybrid methods of scheduling have beenmotivating by the fact that each type of scheduling
technique has its own supremacy and restriction. Therefore, the present work considers the
development of different Hybrid Genetic Algorithms (HGAs) with commonly used list heuristics
for improving the quality of solution in the presence of communication cost.

22
3. PROBLEM STATEMENT
The considered work includes the minimizing the weighted sum of makespan and total
completion time for multiprocessor task scheduling problem. Various assumptions and the fitness
function considered are illustrated below:
3.1 ASSUMPTIONS
• The problem is deterministic with known values of data communication time, task
dependencies & execution time.
• A DAG is used to represent the dependencies along with execution time and
communication cost.
• Communication cost is considered only when the tasks are scheduled on different
processor otherwise it is taken as zero.
• All the processors are homogeneous.i.esame execution time for all the processors.
• Pre-emption of tasks is not allowed.
• Task duplication is not allowed.
• All processors & tasks are accessible at time t = 0.
3.2FITNESSFUNCTION
Fitness function considered in the present work deals with minimizing weighted sum of
makespan and total completion time. Makespan of a schedule is the time at which its last task
completes. Total completion time of a schedule is the summation of completion times of all the
tasks of that schedule. For the requirement of maximum utilization of resources, increase in
throughput, load balancing etc, authors have proposed the bi-criteria decision making fitness
function including the two performance measure for multiprocessor task scheduling which has
been framed as:
Where, F is the bi-criteria fitness function, Cmax is the makespan and Ci is the completion time of
ith task of a schedule with ‘α’ the weight coefficient as per the priority of the performance
measures to be minimised.
4. PROPOSED HYBRID GENETIC ALGORITHM (HGA)
It has been generallyestablished that finding optimality to NP hard problems is not a feasible
option since large amount of computational time is needed for finding of such solutions. In
reality, a superior initial solution can be obtained by a heuristic in arational computational time. A
Genetic algorithm is a valuable population based approach for the multiprocessor task scheduling.
Consequently many researchers [6, 12, 19] have reported success withgenetic algorithm in
achieving good solutions to combinatorial optimization problems. A genetic algorithm starts with
an initial population which can be generated arbitrarily or based on some rules, heuristics and
algorithms. Then in each generation the population follows the sequence of encoding, fitness
evaluation, selection, crossover and mutation until some stopping criteria is met.






−+= ∑=
n
i
iCCMinF
1
max )1( αα

23
Performance of GA depends on the superior chromosomes in initial population for faster
convergence [24]. So, in the present work different list scheduling heuristics (table 1) is used to
obtain a seed sequence which is then combined with (ps-1) randomly generated sequences to give
initial population of size ps. The proposed hybrid genetic algorithm is described below:
Step 1 Encoding
a) Encoding give the illustration of a chromosome. In the present work, chromosome is taken as
(T, P) pair where T is task sequence t1, t2,.......,tn& P is owed processor sequence p1,p2,....,pn.
b) Each task sequence is a variation of task numbers & each processor sequence is a variation of
processor numbers (1, 2... m) with length equal to number of tasks.
Step 2 Initialization
Each task sequence is a variation of task numbers, so each task will be processed according to its
emergence. Each task in the sequence should come out before all of its children and parents due
to dependency. Some mechanism is needed for validating the invalid sequences. The steps for
generation of initial population:
1. Use scheduling heuristic (table 1) for generating the seed sequence.
2. Generate (ps – 1) population arbitrarilyusing the following steps
a) Generate the suitable task sequences (TS) of (Ps-1) using the algorithm as stated
by Bonyadi and Moghaddam[20].
b) Generate the processor sequences (PS) of ps-1arbitrarily.
c) Each task sequence (T) is mapped from TS forarbitrarily selecting the processor
sequence (P) giving each chromosome in the form (T, P) pairsuch that the task
sequence should be followed by mapped processor sequence.
3. Merge the seed sequence with randomly generated population as per population size (Ps).
Step 3 Reproduction
As the task and processor sequences have different nature, so, different reproduction operators are
used for both sequences. Both the sequences are firstly separated from a chromosome and then
used independently for performing crossover and mutation. Task sequences after reproduction
may not follow dependency, so a mechanismstated by Bonyadi and Moghaddam [20] has
beenapplied for validating the task sequences. The valid task sequences after reproduction (TS’)
have mapped to the processor sequences based on minimum fitness value.
The new offspring’s are generated with the following steps:
a) Scores every member of the present population by computing the fitness function
included weighted sum of makespan and total completion time.
b) The individuals as per fitness in the current population act as elite and admit in the next
population.
c) Selects parents for reproduction based on the fitness value as per the selection function.

24
Step 4: Stopping criteria
The algorithm stops when maximum number of generations reaches 100.
4.1 List Scheduling heuristics
List scheduling heuristicsare based on assigning priorities to tasks and then allocating the highest
priority task to processor which gives the minimum start time. The present work uses five list
scheduling heuristics i.e. HLFET,ISH,MCP,ETF,DLS to input one best sequence in initial
population of GA resulting into five different hybrid genetic algorithms HLFET-GA, ISH-GA,
MCP-GA,ETF-GA and DLS-GA. Different heuristics used for the initial seed sequence in the
different hybridgenetic algorithmic are shown in table 1.
Table 1. Different Heuristics for initial feasible solution of different Hybrid GAs
5. RESULTS AND DISCUSSION
The proposed different Hybrid Genetic algorithms (HGA’s) have been implemented in MATLAB
environment. First, a sample multiprocessor task scheduling problem (T18) has beenconsidered
to set the parameters of the proposed GA for the bi-criteria multiprocessor task scheduling
problem. The different parameters setting of GA have been optimised for a sample problem using
design of experiments. Parameters fixed for different proposed and developed HGAs are shown
in table 2.
The value of total completion time is very large as compared to makespan and values of weight
coefficient (α) must be normalized in such a way that minimization of fitness values have
negligible impact on the makespan and total completion time for equal priority. The normalized
values eliminate the effects of certain gross influences.

25
Table 2. Parameters fixed for different HGAs.
There are different methods of normalization and every method requires the best value of the
objectives.Therefore, normalization has been done with the help of different experiments from the
heuristics on the different problems and a value of weight coefficient (α) for a particular problem
are calculated from the relation (irrespective of number of processors) and is given below:
Weight Coefficient (α) = 





+
−
Bestbest
Best
MakespanetiontimeTotalCompl
Makespan
1
The comparative analysis has been done by computing the performance Index (PI) as:-
Performance Index (PI) (%) = 100
lg
1 ×




 −
−
solution
solutionsolution
Best
BestorithmA
Algorithmsolution is the average solution obtained by the different algorithmsand Bestsolution is the
best solution obtained from different algorithmin all the runs. Performance Indexcloser to 100%
provides the good results. Some of the standard problems [12]along with their computed weight
coefficient (α) have been used for the comparative analysis among proposed HGA’s as shown in
table 3.
Table 3.Standard Multiprocessor Task Scheduling Problems along with weight coefficient
All the Hybrid GAs have been run five times for taking final average and comparative analysis
for makespan and total completion time criteria has been shown in fig. 1 and 2 respectively. The
problem T9 and T18 having the variable communication cost and the problems T14_1, T14_2,
T16_1 & T16_2 have the fixed communication cost. It can also be seen that the communication
cost has great impact onthe quality of solution. For the variable communication cost, ISH-GA and

26
ETF-GA provides the best results for the problem T9 and as the task size increase i.e. T18, the
ETF-GA provides the superior results as comparison to others with all the processors for the
makespan and total completion time.
For the problem having fixed communication cost, all the HGA’s are comparable to each others
for all the processors having lesser communication cost. As the communication cost increases
along with task size, all the HGA’s (Except HLFET-GA) provides the similar results. Therefore,
in order to compare the different HGAs, Average Performance Index for all the problems has
been calculated for 2, 3 and 4 processors and shown in table 4.
Average performance index shows the ETF-GA algorithm provides better results for all the
problems considered with average PI of 95.55% and 95.57% for 2 processors, 98.58% and
98.09% for 3 processors and 99.47% and 98.44% for 4 processors for minimization of makespan
and total completion time respectively. Therefore, ETF based HGA when compared to others
provides the best compromise results for the makespan and total completion time criteria for the
larger size multiprocessor task scheduling problems with 2, 3 and 4 processors.
Figure 1. Performance Index (%) of different Hybrid GA for makespan (a) Two Processors (b) Three
Processors (c) Four Processors
.

27
Figure 2. Performance Index (%) of different Hybrid GA for Total Completion Time (a) Two Processors
(b) Three Processors (c) Four Processors
Table 4. Average Performance Index (%) for different HGAs on different processors
MS: Makespan& TCT: Total Completion Time
6. CONCLUSIONS
Present work considers the bi-criteria multiprocessor task scheduling problem on homogeneous
processors with objective of minimizing the weighted sum of makespan and total completion
time. Initial solution from the well known list heuristics have been obtained and combined with
initial population of genetic algorithm to form five different Hybrid Genetic Algorithms (HGA’s)
for the problem. The computation analysis has been done on the some standard benchmark
multiprocessor task scheduling problem with variable and fixed communication cost.
Performance analysis is done with the help of defined performance indexand reveals that ETF
based hybrid GA (ETF-GA) provides finer results for the makespan and total completion time
criteriaparticularlyfor larger and complex problems.

28
REFERENCES
[1] Adam, T.L., Candy, K.M. and Dickson, J., “A Comparison of list scheduling for parallel processing
system”, Communication ACM, Vol. 17, No. 12, pp. 685-690, 1974.
[2] Kruatrachue, B. and Lewis, T.G., “Duplication Scheduling Heuristic, a New Precedence Task
Scheduler for Parallel Systems”, Technical Report, 87-60-3, Oregon State University, 1987.
[3] Hwang, J.J., Chow, Y.C., Anger, F. D. and Lee, C.Y., “Scheduling precedence graphs in systems with
interprocessor communication times”, SIAM Journal of Computing, Vol. 18, No.2, pp. 244–257,
1989.
[4] Wu, M.Y., and Gajski, D.D., “Hypertool A programming aid for message-passing systems”, IEEE
Transactions on Parallel and Distributing System, Vol. 1, No. 3, pp. 330–343, 1990.
[5] Sih, G. C. and Lee, E. A., “A compile-time scheduling heuristic for interconnection- constrained
heterogeneous processor architectures”, IEEE Transactions on Parallel and Distributing System, Vol.
4, No. 2, pp.75–87, 1993.
[6] Hou, E.S.H., Ansari, N. and Hong, R. “A Genetic Algorithm for Multiprocessor Scheduling”, IEEE
Transactions on Parallel and Distributed Systems. Vol. 5, No. 2, pp. 113 – 120, 1994.
[7] Dhodhi, M.K., and Ahmad, I., “ A multiprocessor scheduling scheme using problem-space genetic
algorithms”, Proceedings of IEEE International Conference on Evolutionary Computation, pp. 214–
219, 1995.
[8] Ahmad, I., and Dhodhi, M.K., “Multiprocessor scheduling in a genetic paradigm”, Parallel
compututing, Vol. 22, pp. 395–406, 1996.
[9] Tsuchiya, T., Osada, T., and Kikuno, T., “Genetics-based multiprocessor scheduling using task
duplication”, Journal of Microprocessors and Microsystems, Vol. 22, Vol. 3–4, pp. 197–207, 1998.
[10] Kwok, Y.K., and Ahmad, I., “ Static Scheduling Algorithms for Allocating Directed Task Graphs
to Multiprocessors”, ACM computing surveys, Vol. 31, No. 4, pp. 406-471, 1999.
[11] Zhong, Y.W., and Yang, J.G., “A genetic algorithm for tasks scheduling in parallel multiprocessor
systems”, Proceedings of the Second International Conference on Machine Learning and Cybernetics,
pp.1785–1790, 2003.
[12] Wu, A.S., Yu, H., Jin,S., Lin, K.C and Schiavone,G., “An incremental genetic algorithm approach to
multiprocessor scheduling”, IEEE Transactions on Parallel and Distributed Systems, Vol. 15, No. 9,
pp. 824–834, 2004.
[13] Sivanandam, S.N.., Visalakshi, ,P., and Bhuvaneswari, A.,”Multiprocessor Scheduling Using Hybrid
Particle Swarm Optimization with Dynamically Varying Inertia “,International Journal of Computer
Science & Applications, Vol. 4, No.3, pp 95-106, 2007.
[14] Yoo,M., and Gen, M., “Scheduling algorithm for real-time tasks using multiobjective hybrid genetic
algorithm in heterogeneous multiprocessors system”, Computers and Operations Research, Vol. 34,
pp. 3084 – 3098, 2007.
[15] Dahal, K., Hossain, A. , Varghese, B. , and Abraham, A., “ Scheduling in Multiprocessor System
Using Genetic Algorithms” Proceedings Of 7th Conference on Computer Information Systems and
Industrial Management Applications(CISIM-2008), pp. 281 – 286, 2008.
[16] Azghadi M. R., Bonyadi, M.R., Hashemi, S. and Moghadam, M.E, “A Hybrid Multiprocessor Task
Scheduling Method Based on Immune Genetic Algorithm”, Proceedings of 5th International ICST
Conference on Heterogeneous Networking for Quality, Reliability, Security and Robustness Hong
Kong, Hong Kong , July 28 - 31, 2008.
[17] Jouglet, C. O˘guz, and M. Sevaux, “Hybrid Flow-Shop: a Memetic Algorithm Using Constraint-
Based Scheduling for Efficient Search”, Journal of mathematical modeling and algorithms, Vol. 8, pp.
271 -292, 2008.
[18] Hwang, R., Gen, M. and Katayam, H., "A comparison of multiprocessor task scheduling algorithms
with communication costs," The Journal of Computers & Operations Research, vol. 35, pp. 976 –
993, 2008.
[19] Kim, E.S., Sung, C.S. and Lee, I. S., "Scheduling of parallel machines to minimize total completion
time subject to s-precedence constraints," The Journal of Computers & Operations Research, Vol. 36,
pp. 698 - 710,2009.

29
[20] Bonyadi, M.R and Moghaddam, M. E, “A bipartite genetic algorithm for multi-processor task
scheduling” , International Journal of Parallel Programming, Vol. 37, No. 5, pp. 462- 487, 2009.
[21] Goh, C. K. , Teoh, E. J., and Tan, K. C., “A hybrid evolutionary approach for heterogeneous
multiprocessor scheduling”, Soft Computing, Vol.13, pp. 833–846, 2009.
[22] Wen,Y., Xu, H., and Yang, J., “A heuristic-based hybrid genetic-variable neighborhood search
algorithm for task scheduling in heterogeneous multiprocessor system”, Information Sciences Vol.
181, pp. 567–581, 2011.
[23] Mohamed, M. R., and Awadalla, M. H. A., “Hybrid Algorithm for Multiprocessor Task Scheduling”,
IJCSI International Journal of Computer Science Issues, Vol. 8, Issue 3, No. 2, 2011
[24] Roy, P., Alam, M.M., and Das, N., “Heuristic Based Task Scheduling in Multiprocessor Systems
With Genetic Algorithm By Choosing The Eligible Processor”, International Journal of Distributed
and Parallel Systems (IJDPS) Vol.3, No.4, 2012.

HYBRID GENETIC ALGORITHM FOR BI-CRITERIA MULTIPROCESSOR TASK SCHEDULING WITH COMMUNICATION DELAY

More Related Content

What's hot (17)

Viewers also liked (20)

Similar to HYBRID GENETIC ALGORITHM FOR BI-CRITERIA MULTIPROCESSOR TASK SCHEDULING WITH COMMUNICATION DELAY (20)

Recently uploaded (20)

HYBRID GENETIC ALGORITHM FOR BI-CRITERIA MULTIPROCESSOR TASK SCHEDULING WITH COMMUNICATION DELAY