EA-MSCA: An effective energy-aware multi-objective modified sine-cosine algorithm for real-time task scheduling in multiprocessor systems: Methods and analysis
Introduction
Until recently, all the computing devices were based on a single processor system-on-chip (SPS) which limited their computing capabilities to take a long time with low throughput which became a considerable obstacle to growth, especially for industry and technology applications (Hadizadeh and Tanghatari, 2017, Razian and MahvashMohammadi, 2017). The multiprocessor system-on-chip (MPS) alleviated many of those problems by building a number of small processors known as processing cores, or also called processing elements, on a single chip to allow processing in parallel for more than one task at a time to reduce the delay time and improve throughput. MPS platforms can be classified to heterogeneous and homogenous MPS. In a heterogeneous MPS platform, the processing cores have various instruction set architecture (ISA). On the contrary, the homogenous platform has processing cores with the same ISA. If the processing cores in the homogenous platform are totally identical with respect to performance and power, this platform is known as a symmetric MPS platform. Rather than, it is called an asymmetric MPS platform (Yun, Hwang, & Kim, 2019).
MPS is a great boon to real-time applications, such as control systems, radar tracking, and audio/video streaming (Mo, Kritikakou, & Sentieys, 2018) in providing timely responses in applications that cannot tolerate delay. Despite their advantages in speed and throughput, MPS consumes very large amounts of energy which increases the cost of their operation. Consequently, decreasing the energy consumption of MPS is highly desirable to save costs and battery lifetime, as well as to improve the performance MPS.
To reduce the energy consumed by MPS, the tasks assigned to them must be accurately scheduled; this process is known as task scheduling (TS) and the problem is known as the task scheduling MPS (TSMPS) problem. Unlike MPS, TS on SPS is straightforward because the tasks are executed sequentially and the dependency among tasks is respected if it is detected. MPS, however, must distribute and arrange the tasks while taking into consideration the dependency among processors. In addition, among the processors, there is a communication cost to exchange the data regarding the related tasks allocated to them. Although MPS can implement tasks in parallel at the same time to increase throughput and minimize the time taken, energy consumption would increase. To overcome this drawback, dynamic energy and frequency scaling (DVFS) can be used to reduce energy consumption (Wang, Qian, Yuan, & You, 2017).
The tasks assignment to the processing elements in MPS is considered a challenging problem that is NP-hard (Balin, 2011, Engin et al., 2011). The traditional methods, such as exhaustive search, proposed to overcome this problem suffer from time complexity and increased memory space, which motivates researchers to use meta-heuristic algorithms to overcome those disadvantages. In addition, in comparison with the heuristic algorithms, meta-heuristic algorithms could substantially avoid the local minima, in addition to alleviating using domain-specific knowledge and therefore they could be applied to solve several other problems. Meta-heuristic algorithms are ideal in that regard due to their ability to obtain the near-optimal solution for many optimization problems at a reasonable time (Abdelhafez et al., 2019, Jain et al., 2019, Mahato and Singh, 2018, Hashemi and Rahmani, 2018, Mafarja and Mirjalili, 2018, Mirjalili et al., 2018, Abdel-Basset et al., 2020, Abdel-Basset et al., 2020). Several meta-heuristic algorithms have been proposed to improve the performance in addressing TSMPS-the next section reviews the major work.
Some of the papers reviewed in the next section were applied to the task graphs with small task sizes up to 230 and this didn’t strengthen the motivation for using those algorithms with increasing the size of the tasks since their performance is unknown so-far. The remainder of the papers report approaches still suffer from becoming trapped in local minima that prevent them from reaching better outcomes.
Recently, a population-based algorithm called the sine-cosine algorithm (SCA) (Mirjalili, 2016) has been proposed for solving the uni-modal, multi-modal, and composite test functions (Digalakis and Margaritis, 2001, Yao et al., 1999). The advantages of the SCA that motivate this study are that it is simple, easy to implement, and easy to be modified to improve its performance. Despite the significant success of SCA in solving several real problems, it still suffers from the following main disadvantage in its optimization core: it searches for a better solution around the current solution with a step size located between the current solution and the best solution multiplied by a random number ranging of 0 and 2, which may make some regions intractable for SCA, or may reduce the convergence speed toward the optimal solution because of multiplying the best-so-far solution by a random number. This disadvantage therefore motivates the proposal of a multi-objective modified sine-cosine algorithm (EA-M2SCA) by dividing the SCA optimization process into three stages:
- •
The first stage utilizes the start of the optimization process to explore the regions of the search space of the problem as much as possible.
- •
The second utilizes some of the optimization processes to explore the regions around a solution selected randomly from the population.
- •
The last stage moves the current solution toward the best-so-far solution in order to accelerate the convergence toward the optimal solution.
- •
A proposed modification to multi-objective SCA (EA-M2SCA) to improve its performance when solving the TSMPS.
- •
To improve the performance of EA-M2SCA, it was integrated with the polynomial mutation strategy in two effective manners to avoid stuck into local minima with increasing the convergence toward the best-so-far, this version was named as EA-MHSCA.
- •
EA-MHSCA and EA-M2SCA are compared with a number of well-known multi-objective algorithms to check its efficacy.
The remainder of this paper is organized as follows: Section 2 provides a review of relevant literature, Section 3 describes MPS and performance measures, Section 4 presents our proposed approach, Section 5 Experimental settings, 6 Results and discussion describe the experimental settings and the results and Section 7 provides conclusions about this work in addition to introducing a number of future works.
Section snippets
Literature review
Several algorithms have been proposed to reduce the energy consumed by MPS, especially with the significant growth of the MPS. In Abdel-Basset et al. (2020), an approach known as energy and memory-aware retiming conditional task graph (EMRCTG) was proposed to use task-level coarse-grained software pipelining with DVFS to reduce the energy consumption while satisfying memory capacity constraints. Qin, Zeng, Kurachi, Matsubara, and Takada (2019) proposed an approach to minimize the energy
Problem formulation and Performance measures
In this section, the problem formulation for the TSMPS, in addition to the performance metrics are discussed in detail.
The proposed solution approach
This section illustrates the steps of the proposed algorithms: hybrid modified sine cosine algorithm (HMSCA) and modified sine-cosine algorithm (MSCA) for tackling the TSMPS: initialization, performance measure, modified sine-cosine algorithm, and load balancing improvement strategy. As mentioned before, this problem is dealt with as a multi-objective problem and the obtained solutions will be evaluated using the Pareto dominance to find the best solutions. Additionally, to find several
Experimental settings
In this section, the parameter settings of the algorithms besides the datasets used within the next experiments will be discussed. All the algorithms in our experiments are executed in java and set up on a machine having Intel®Core™i7-4700MQ CPU @ 2.40 GHz, 16 GB of RAM, and prepared with Windows 10 platform. Dataset description and parameter tunings are highlighted in the following subsections.
Results and discussion
This section describes the results of the experiments comparing the proposed algorithm with the recent robust algorithms mentioned in the previous section.
Conclusion and future work
In this paper, the multi-objective sine-cosine algorithm (MSCA) is modified to address the task scheduling for MPS as a multi-objective optimization problem using the Pareto optimality. The modified MSCA (M2SCA) is divided into three stages. The first stage explores the search space of the problem to minimize the probability of falling into local minima. The second searches around a solution selected randomly to avoid becoming trapped into local minima. The third stage searches around the
CRediT authorship contribution statement
Mohamed Abdel-Basset: Investigation, Methodology, Resources, Visualization, Software, Writing - original draft, Writing - review & editing. Reda Mohamed: Investigation, Methodology, Resources, Visualization, Software, Writing - original draft, Writing - review & editing. Mohamed Abouhawwash: Conceptualization, Methodology, Resources, Visualization, Writing - review & editing. Ripon K. Chakrabortty: Conceptualization, Methodology, Writing - review & editing. Michael J. Ryan: Investigation,
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
aaa
References (58)
- et al.
Performance analysis of synchronous and asynchronous distributed genetic algorithms on multiprocessors
Swarm and Evolutionary Computation
(2019) - et al.
Symbiotic organism search optimization based task scheduling in cloud computing environment
Future Generation Computer Systems
(2016) - et al.
Enhanced particle swarm optimization for task scheduling in cloud computing environments
Procedia Computer Science
(2015) Non-identical parallel machine scheduling using genetic algorithm
Expert Systems with Applications
(2011)- et al.
An intelligent/cognitive model of task scheduling for iot applications in cloud computing environment
Future Generation Computer Systems
(2018) - et al.
A diversity indicator based on reference vectors for many-objective optimization
Information Sciences
(2018) - et al.
Priority-based and conflict-avoidance heuristics for multi-satellite scheduling
Applied Soft Computing
(2018) - et al.
Omni-optimizer: A generic evolutionary algorithm for single and multi-objective optimization
European Journal of Operational Research
(2008) - et al.
Modified heft algorithm for task scheduling in cloud environment
Procedia Computer Science
(2018) - et al.
An efficient genetic algorithm for hybrid flow shop scheduling with multiprocessor task problems
Applied Soft Computing
(2011)
A novel nature-inspired algorithm for optimization: Squirrel search algorithm
Swarm and Evolutionary Computation
An improved hybrid quantum-inspired genetic algorithm (hqiga) for scheduling of real-time task in multiprocessor system
Applied Soft Computing
A multi-objective quantum-inspired genetic algorithm (mo-qiga) for real-time tasks scheduling in multiprocessor environment
Procedia Computer Science
Whale optimization approaches for wrapper feature selection
Applied Soft Computing
On maximizing reliability of grid transaction processing system considering balanced task allocation using social spider optimization
Swarm and Evolutionary Computation
Evaluation framework for energy-aware multiprocessor scheduling in real-time systems
Journal of Systems Architecture
Sca: a sine cosine algorithm for solving optimization problems
Knowledge-based systems
Task allocation, migration and scheduling for energy-efficient real-time multiprocessor architectures
Journal of Systems Architecture
Pareto-optimal solutions for multi-objective optimization of fed-batch bioreactors using nondominated sorting genetic algorithm
Chemical Engineering Science
A hybrid algorithm for task scheduling on heterogeneous multiprocessor embedded systems
Applied Soft Computing
Adaptive genetic algorithm for energy-efficient task scheduling on asymmetric multiprocessor system-on-chip
Microprocessors and Microsystems
Thermal-aware correlated two-level scheduling of real-time tasks with reduced processor energy on heterogeneous mpsocs
Journal of Systems Architecture
A novel equilibrium optimization algorithm for multi-thresholding image segmentation problems
Neural Computing and Applications
Energy-aware marine predators algorithm for task scheduling in iot-based fog computing applications
IEEE Transactions on Industrial Informatics
A genetic algorithm inspired task scheduling in cloud computing
A fast and elitist multiobjective genetic algorithm: Nsga-ii
IEEE Transactions on Evolutionary Computation
Energy-aware task scheduling on heterogeneous computing systems with time constraint
IEEE Access
Cited by (41)
METSM: Multiobjective energy-efficient task scheduling model for an edge heterogeneous multiprocessor system
2024, Future Generation Computer SystemsMultiprocessor task scheduling using multi-objective hybrid genetic Algorithm in Fog–cloud computing
2023, Knowledge-Based SystemsAn adaptive quadratic interpolation and rounding mechanism sine cosine algorithm with application to constrained engineering optimization problems
2023, Expert Systems with ApplicationsCitation Excerpt :For example, Krishna Priya, Deepalakshmi, and Saravana Selvam (2021) applied an improved SCA algorithm to multimedia content distribution in the cloud environment. Abdel-Basset et al. (Abdel-Basset, Mohamed, Abouhawwash, Chakrabortty, & Ryan, 2021) applied SCA to handle multi-objective problems to solve real-time task scheduling in multiprocessor systems. Khrissi et al. (Khrissi, 2022) applied SCA in combination with clustering methods to the field of image segmentation.
Improved sine algorithm for global optimization
2023, Expert Systems with ApplicationsCitation Excerpt :The algorithm was applied to various engineering problems, such as the design of springs. Abdel-Basset, Mohamed, Abouhawwash, Chakrabortty, and Ryan (2021) developed a multi-objective SCA for the task scheduling of multiprocessor systems by implementing a polynomial mutation strategy in two different ways during the search phase to avoid premature convergence. In addition, they also adopted different random selection schemes to prevent the algorithm from finding a local optimal value.
A multi-strategy improved slime mould algorithm for global optimization and engineering design problems
2023, Computer Methods in Applied Mechanics and EngineeringCitation Excerpt :In recent years, MAs have been successfully used to address many real-world optimization problems such as structural damage detection problems [2–8], feature selection [9], and solar radiation modelling problems [10]. These methods mainly include genetic algorithm (GA) [11], particle swarm optimization (PSO) [12–15], moth-flame optimization (MFO) [3,16], grey wolf optimizer (GWO) [4,17,18], sine cosine algorithm (SCA) [19–23], salp swarm algorithm (SSA) [5,24], multi-verse optimizer (MVO) [6,25], arithmetic optimization algorithm (AOA) [26], et al. On the basis of the no-free-lunch (NFL) theorem [27], although the conventional MAs are widely used, they cannot achieve the global optimal solution for all optimization issues.
A novel algorithm for priority-based task scheduling on a multiprocessor heterogeneous system
2022, Microprocessors and MicrosystemsCitation Excerpt :Then, heuristic methods became popular as they could be used to solve task scheduling problems with partial information in polynomial time [11–13,43,44]. Recently, metaheuristic algorithms have become very popular in this area of research due to their randomized search procedure ([16–19,24,25,45,46]; Basset et al., [66,67]). Metaheuristic algorithms produce better task schedules than heuristic algorithms.