An Investigation of Optimal Project Scheduling and Team Staffing
in Software Development using Particle Swarm Optimization
Simos Gerasimou
1
, Constantinos Stylianou
2
and Andreas S. Andreou
1
1
Department of Computer Engineering and Informatics, Cyprus University of Technology,
31 Archbishop Kyprianou Ave., P.O. Box 50329, Lemesos, 3036, Cyprus
2
Department of Computer Science, University of Cyprus,
75 Kallipoleos Ave., P.O. Box 20537, Lefkosia, 1678, Cyprus
Keywords: Software Project Management, Project Scheduling, Team Staffing, Particle Swarm Optimization.
Abstract: Software development organizations often struggle to deliver projects on time, within budget and with the
required quality. One possible cause of this problem is poor software project management and, in particular,
inadequate project scheduling and ineffective team staffing. This paper investigates the application of a
particle swarm optimization algorithm to help software project managers perform these activities
effectively. Specifically, the proposed approach aims to create optimal project schedules by specifying the
best sequence for executing a project’s tasks and minimizing the total project duration. Simultaneously, it
seeks to form skilful and productive working teams with the best utilization of developer skills. These
considerations have been suitably encoded into the algorithm, with several hard constraints and objective
functions appropriately formulated so as to assess the generated solutions with respect to their feasibility
and also their quality. The initial results obtained are quite encouraging for the majority of the performed
tests and indicate that the proposed approach is able to deal with the issues of scheduling and staffing in
software project management.
1 INTRODUCTION
One of the serious problems concerning the majority
of software development organisations is the high
rate of software project failures. According to the
Standish Group’s CHAOS Report of 2009, only
32% of projects produced software systems that
were delivered successfully on time and within
budget and also provided the required features and
functionality (Standish Group, 2009). These figures,
give strong indications that software development
companies systematically fail to accurately plan and
properly measure their development processes, and
the reasons leading to low success rates have,
therefore, been the focal point of many software
engineering researchers.
Among the most significant causes attributed to
software project failures has been the insufficient
and inappropriate practices followed by software
project managers regarding project scheduling and
team staffing activities. In the former case, incorrect
estimates both before and during software
development have been found to play a crucial role
in software project delays and overruns, whereas in
the latter case, assigning project tasks to less suitable
project team members is one of the main causes of
low quality end-products.
The research presented in this paper is an initial
investigation to deal with these issues of software
project management through a swarm intelligence
approach that facilitates both the scheduling of
project tasks and the allocation of the most suitable
team members to tasks in an automated way.
Specifically, the approach targets two goals. Firstly,
to construct an optimal sequence of task executions
and to help minimize software project duration
without any violation of possible dependencies
existing between tasks. Secondly, to form an
efficient and operational software project team with
the best possible utilization of skills measured in
terms of developer experience.
2 RELATED WORK
There have been a number of approaches proposed
168
Gerasimou S., Stylianou C. and S. Andreou A..
An Investigation of Optimal Project Scheduling and Team Staffing in Software Development using Particle Swarm Optimization.
DOI: 10.5220/0004001001680171
In Proceedings of the 14th International Conference on Enterprise Information Systems (ICEIS-2012), pages 168-171
ISBN: 978-989-8565-11-2
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
over the years that aim at helping software project
managers decide on various technical factors such as
project duration and effort as well as developer
availability, with most of the techniques proposed
tackling scheduling and staffing as an optimization
problem. Μany researchers have focused on using
techniques found in the area of computational
intelligence, as these have been proven to be
extremely efficient for solving real-world problems
that are large in size and high in complexity. The
most common techniques include evolutionary
algorithms (Alba and Chicano, 2007; Chang et al.,
2008; Ren, Harman and Di Penta, 2011), fuzzy logic
(Callegari and Bastos, 2009) and constraint
satisfaction (Barreto, Barros and Werner, 2008).
These have been adopted mainly due to their
abilities to reduce problem search spaces and to
model complex problems where there is a lack of
mathematical analysis, as well as to effectively
handle NP-hard problems (Chang et al., 2008).
The attempt presented here takes into account the
non-interchangeable nature of human resources and
aims to optimize assignments so that the level of
developer experience is fully utilized, thus
promoting quality software systems. Furthermore,
swarm intelligence is investigated as a means to
perform the optimization.
3 PROPOSED METHODOLOGY
3.1 Representation and Encoding
A software project comprises a number of tasks that
must be performed in a predetermined sequence,
with the dependencies between them satisfied at all
times. Each task has a specified duration and
requires developers to possess a set of skills in order
to perform it. It is a project manager’s responsibility
to construct the project and form the development
team and, in order to help project managers achieve
this, a particle swarm optimization (PSO) algorithm
was adopted. PSOs are a computational intelligence
technique inspired by biological evolution occurring
in nature (Eberhart and Kennedy, 1995). Swarm
particles denote a candidate solution to the problem,
and in this attempt the same representation defined
in Stylianou and Andreou (2011) is used. Each
particle’s dimension uses mixed-type encoding to
hold scheduling information and the assigned
developers. Scheduling information is expressed by
each task’s starting day and the team staffing
information is represented by a binary vector, where
each bit shows whether or not a developer has been
assigned to a task.
3.2 Particle Evaluation
The evaluation of each candidate solution
is
assessed based on two factors, as shown in Eq. (1):
(a) the computation of the degree of satisfaction of
hard constraints and (b) the calculation of its fitness
using objective functions.
=
+
(1)
where
and
denote the
computed values regarding the hard constraints and
objective functions, respectively. The first factor is
used to assess the feasibility of a solution, whereas
the second factor shows its quality.
A candidate solution is considered feasible if and
only if it satisfies the imposed constraints, as shown
in Eqs (2)-(4). Each constraint contains a penalty
coefficient with a negative value in order to stress
the existence of a violation.
1
=
×#_
×#
(2)
2
=
×
#_
#_
(3)
3
=
×
#_
#_
(4)
Constraint 1 measures if there are violations of
task dependencies, since it is required that each
task’s starting day must be set after all of its
predecessors have completed. Constraint 2
measures if all skills required by a task are fulfilled
by the developers assigned, since if the team does
not possess one or more required skills then the task
will not complete successfully and defects could
occur. Constraint 3 measures if conflicts arise when
developers are assigned to tasks, as they are not
permitted to work on more than one task at any
given time. The final constraint value of a particle is
the summation of the individual constraint terms.
The fitness of a solution is evaluated using the
two objective functions in Eqs (5) and (6). The
former considers the duration of the project and the
latter takes into account the experience of the
assigned developers.
=
∑
(5)
=
∑
∑
(6)
The objective function
aims to schedule
AnInvestigationofOptimalProjectSchedulingandTeamStaffinginSoftwareDevelopmentusingParticleSwarm
Optimization
169
tasks so that there are no needless (idle) delays
within the project, thus minimizing its overall
duration. On the other hand, objective function
aims to ensure that the teams will be the most
suitable for the accomplishment of each task, and
uses each assigned developer’s degree of experience
in the skills required. Since the two objectives are
directly competing, it is often likely that the
algorithm’s attempt to increase one objective would
cause the other to lower. Therefore, a trade-off
mechanism using weights for each objective
function, shown in Eq. (7), was implemented to
allow software project managers to decide which of
the two objectives is more significant for them.
=
×
+
×
(7)
where 0<
,
<1 and
+
=1.
4 EXPERIMENTAL RESULTS
4.1 Design of Experiments
Initially, a survey was conducted with a number of
software development SMEs in Cyprus in order to
find out the driving factors influencing the size and
complexity of a software project. With the
information obtained, a total of 7 projects of varying
size and complexity were used aiming to represent
real-world software project case studies. The factors
taken into account and their respective values in
each project are provided in Table 1. Furthermore,
three different sets of ratios for the weight values
and
(Eq. (7)) were used: equal importance (1:1),
importance to project scheduling (9:1) and
importance to developer experience (1:9).
4.2 Parameters and Execution
A combination of Constriction-PSO and Binary-PSO
(Poli, Kennedy and Blackwell, 2007) variations
were selected as the most suitable. Also, due to the
multimodal nature of the problem having many
global/local minimum, a low-connected ring
topology was used so the swarm could adequately
examine the search space and avoid premature
convergence in local optimal solutions. The swarm
size was kept constant at 60 particles and all 7
projects were executed 10 times for each weight
ratio variation, with a maximum 10
6
number of
iterations. In case that stagnation was observed, a
partial re-initialization of positions and velocities
took place. Finally, the penalty values for the
constraints in Eqs. (2)-(4) were specified to -100.
4.3 Results and Discussion
As previously mentioned, the primary objective of
this research attempt is to carry out an initial
investigation as to whether the proposed approach
produces acceptable solutions within the context of
software project management. Therefore, each
particle in the swarm was assessed, firstly, based on
whether it represents a feasible software project
schedule and developer assignments and, secondly,
based on its ability to generate optimal solutions.
The results of the executions are presented in Table
2. For the first project, all the final particles at the
end of the algorithm’s executions represent feasible
solutions (since its feasibility rate equals 100%) and
in addition all of them are optimal solutions (with a
100% hit rate). As the complexity and size of the
software projects increase however, these
percentages begin to decrease. Despite this, the
algorithm always generates solutions that are
feasible (but not necessarily optimal) even in the
most complex and difficult project instances (i.e., 5
to 7). This indicates that the algorithm is highly
capable of constructing adequate solutions with
respect to the hard constraints imposed.
With respect to the quality of the produced
solutions, the hit ratios in Table 2 show that the
algorithm performs sufficiently well with the first
four projects for all weight ratio variations. Here, the
hit ratio percentages reach a maximum value of
Table 1: Software projects used to study the particle swarm optimization algorithm.
Project
Number of
Tasks
Number of
Dependencies (Rate)
Average Number of
Skills per Task
Number of
Available Developers
Average Number of
Skills per Developer
1 10 13 (29%) 2 10 2
2 14 16 (18%) 2 10 1.5
3 18 24 (16%) 2 10 1.2
4 18 24 (16%) 2 5 0.7
5 25 15 (5%) 2.5 8 1
6 30 62 (14%) 3.3 18 2
7 30 62 (14%) 3.3 10 1
ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems
170
Table 2: Average feasibility and hit ratio percentages for each project for each weight ratio variation.
Weight
Ratios
Average Feasibility Rate (%) | Hit Ratio (%)
1 2 3 4 5 6 7
1:1 100 | 100 99.1 | 30.0 97.9 | 50.0 96.6 | 30.0 89.0 | 0.0 88.2 | 0.0 85.0 | 0.0
9:1 100 | 100 99.6 | 50.0 97.8 | 40.0 95.6 | 30.0 88.8 | 0.0 87.3 | 0.0 83.8 | 0.0
1:9 100 | 100 98.0 | 90.0 96.0 | 80.0 95.0 | 70.0 90.0 | 0.0 89.0 | 0.0 87.0 | 0.0
100% in the first project but as the complexity
increases, a progressive decrease is observed
reaching as low as 30% in the fourth project. A
possible explanation for the behaviour of the
algorithm is that it encounters more difficulties when
trying to satisfy the constraints since, intuitively, the
fewer the number of available developers, the more
likely that assignment conflicts will arise. With
regards to projects 5 to 7, the algorithm experiences
some difficulties in finding optimal solutions,
despite being able to frequently generate feasible
solutions (within 80%-90% of the time). This can
suggest that the large increase in the complexity and
size of software projects causes difficulties in the
evolution of the algorithm and consequently to the
generation of optimal solutions.
5 CONCLUDING REMARKS
The results obtained from various executions of the
algorithm indicated that PSO is a promising
approach for software project scheduling and team
staffing, which performs sufficiently well in the
majority of the projects examined in this paper. The
average feasibility ratio of the solutions generated is
more than 83% proving that most of the particles in
a swarm reside in feasible search space area.
However, some difficulties were encountered in the
cases with larger-sized and more complex software
projects, where the number of tasks, the type of
dependencies and the number of available
developers were shown to influence the ability of the
algorithm to produce optimal solutions. Specifically
in certain instances, the existence of “needless” gaps
in project schedules was observed, despite satisfying
all constraints. In order to increase the quality of
solutions, an adjustment can be made to the
objective functions so that they can more adequately
handle gaps or by introducing new objective
functions that could assist the swarm during its
evolution. Furthermore, due to the obvious
conflicting nature of the present objective functions,
an implementation of a multi-objective version of
the algorithm may perhaps be able to produce better
results. These abovementioned adjustments are
scheduled for future work along with
experimentation with real software projects, which is
currently in process with the collaboration of local
software SMEs for the provision of data.
REFERENCES
Alba, E. and Chicano, J. F., 2007. Software project
management with GAs. Inform. Sciences, 177(11), pp.
2380-2401.
Barreto, A., Barros, M. d. O. and Werner, C. M. L., 2008.
Staffing a software project: A constraint satisfaction
and optimization-based approach. Comput. Oper. Res.,
35(10), pp. 3073-3089.
Callegari, D. A. and Bastos, R. M., 2009. A multi-criteria
resource selection method for software projects using
fuzzy logic. In 11th International Conference on
Enterprise Information Systems. Milan, Italy, 6-10
May 2009. Berlin, Germany: Springer-Verlag.
Chang, C. K., et al., 2008. Time-line based model for
software project scheduling with genetic algorithms.
Inform. Software Tech., 50(11), pp. 1142-1154.
Eberhart, R. and Kennedy, J., 1995. A new optimizer
using particle swarm theory. In 6th International
Symposium on Micro Machine and Human Science.
Nagoya, Japan, 4-6 October 1995. Piscataway, NJ,
USA: IEEE Industry Applications Society.
Poli R., Kennedy J., Blackwell T., 2007. Particle swarm
optimization: an overview. Swarm Intelligence, 1(1),
pp. 33–57.
Ren, J., Harman, M. and Di Penta, M., 2011. Cooperative
co-evolutionary optimization of software project staff
assignments and job scheduling. In International
Symposium on Search Based Software Engineering.
Szeged, Hungary, 10-12 September 2011. Berlin,
Germany: Springer-Verlag.
Standish Group, 2009. Standish Group CHAOS Report.
Boston, MA, USA: Standish Group International, Inc.
Stylianou, C and Andreou, S. A. 2011. Intelligent
Software Project Scheduling and Team Staffing with
Genetic Algorithms. In 7th IFIP Conference on
Artificial Intelligence Applications and Innovations,
Corfu, Greece, 15-18 September 2011, Berlin,
Germany: Springer-Verlag.
AnInvestigationofOptimalProjectSchedulingandTeamStaffinginSoftwareDevelopmentusingParticleSwarm
Optimization
171
