Meta-schedulers for grid computing based on multi-objective swarm algorithms

doi:10.1016/j.asoc.2012.12.030

Applied Soft Computing

Volume 13, Issue 4, April 2013, Pages 1567-1582

https://doi.org/10.1016/j.asoc.2012.12.030 Get rights and content

Abstract

Job scheduling is a challenging task on grid environments because they must fulfill user requirements. Scientists often have deadlines and budgets for their experiments (set of jobs). But these requirements are in conflict with each other – cheaper resources are slower than the expensive ones. In this paper, we have implemented two multi-objective swarm algorithms. One of them is based on a biological behavior – Multi-Objective Artificial Bee Colony (MOABC) – and the other on physics – Multi-Objective Gravitational Search Algorithm (MOGSA). Multi-objective properties enhance the optimization of execution time and cost per experiment. These algorithms are evaluated regard to the standard and well-known multi-objective algorithm – Non-dominated Sorting Genetic Algorithm II (NSGA II) – in order to prove the goodness of our multi-objective proposals. Moreover, they are compared with real meta-schedulers as the Workload Management System (WMS) from the most used European grid middleware, gLite, and the Deadline Budget Constraint (DBC) from Nimrod-G, that takes into account the same requirements. Results show us that MOABC offers better results in all the cases using diverse workflows with dependent jobs over different grid environments.

Graphical abstract

Highlights

► We have implemented two multi-objective swarm algorithms, MOABC and MOGSA, for grid scheduling. ► These algorithms are evaluated regard to the standard and well-known multi-objective algorithm NSGA II. ► They are compared with real meta-schedulers as WMS and DBC. ► The experiments are carried out over different grid environments using diverse workflows with dependent jobs. ► MOABC offers better results in all the cases for both objectives, execution time and cost.

Introduction

The paradigm of grid computing is defined as a parallel and distributed system that allows to share, select and collect autonomous resources geographically distributed in a dynamic way. All these actions are carried out in execution time depending on the availability, capacity, cost and quality of the resources required by the users. This focus emerges from the synergy between the cooperation among computing resources with a decentralized control and providing them as services. Therefore, one of the most important and challenging actors, that participate in this decentralized control, are the meta-schedulers, also known as resource brokers. This service is implemented in the middleware which facilitates the grid environment management. The main function of a meta-scheduler is to assign the jobs to adequate resources following computational requirements and quality of service demanded by the users.

Grid computing is widely used in the scientific world solving complex experiments (set of interdependent jobs) that require a high performance. Scientists often have to consider deadlines and budgets of their experiments related to important projects. Because of that, the optimization of execution time and cost is a key factor to consider in the job scheduling process carried out by the meta-schedulers. However, these types of objectives are in conflict each other, because usually cheaper resources are slower than expensive ones, hence a multi-objective optimization is required.

In this paper, a study of bio-inspired algorithms is presented to solve this multi-objective problem. The implemented algorithms are based on Swarm Intelligence (SI). Swarm intelligence is a kind of intelligence that emerges from the collaboration and competition among individuals. In particular, two novel swarm algorithms from different fields – biology and physics – have been adapted and evaluated. Multi-Objective Artificial Bee Colony (MOABC) is based on the Artificial Bee Colony [1], [2] from the biological field and its collaborating agents are represented as bees. Multi-Objective Gravitational Search Algorithm (MOGSA) is built from the Gravitational Search Algorithm [3] and the planets are its agents, according to the physical field. One of the main contributions of this research is the adaption of these algorithms to deal with multi-objective requirements that are in conflict each other (execution time and cost). Therefore, to give more reliability to this multi-objective study, an evaluation with a well-known standard multi-objective algorithm – Non-dominated Sorting Genetic Algorithm II (NSGA II) [4] – has been accomplished. GridSim [5] is the simulator used to implement all the meta-schedulers.

GridSim³ is a Java-based toolkit for modelling and simulating distributed resource management in Grid environments. GridSim is based on SimJava, a general-purpose discrete-event simulation package implemented in Java. All components in GridSim communicate with each other through message passing operations defined by SimJava. It allows modelling of heterogeneous types of resources and the resources can be modeled operating under space or time shared mode. The resource capability can be defined in the form of MIPS (Million Instructions Per Second) and they can be located in any time zone. Moreover, applications with different parallel application models can be simulated. GridSim toolkit is suitable for application scheduling simulations in Grid Computing environment. GridSim is of great value to test new algorithms and strategies in a controlled environment. By using GridSim, it is possible to perform repeatable experiments and studies that are not possible in a real dynamic Grid environment. The main advantage of GridSim is that various allocation or scheduling policies can be implemented and integrated into GridSim easily, by extending them from one of the classes. Research students in the GRIDS Laboratory⁴ are themselves heavy users of GridSim and extend it whenever necessary for their own research needs. In the last 5 years, GridSim has been continuously extended in this manner to include many new capabilities and has also received contributions from external collaborators. Therefore, it has been chosen due to these advantages and overall to offer the facilities to configure complex topologies and resource features such as processing speed, MIPS (Million Instructions Per Second), or cost of resources per time unit. Furthermore, thanks to the GridSim flexibility, it has been modified to support workflows with dependent jobs, due to the importance to control the execution time in this type of workflows (child jobs have to wait until their parents are successfully executed). In particular, in this research six workflows – Gaussian, Gauss-Jordan, LU decomposition, Find-Max, Fast Fourier Transform and Stencil – have been tested with all the implemented meta-schedulers to study their behaviour in each situation. In addition, two different grid environments have been used to reinforce the study of their behaviour in different scenarios. Also, the best meta-scheduler, in this case the proposal MOABC, has been compared with two real meta-schedulers: the Workload Management System (WMS) and the Deadline Budget Constraint meta-scheduler (DBC) to show the relevance of our results.

This paper is organized as follows. Section 2 presents the related work. Section 3 exposes the problem including an introduction of the multi-objective approach. Section 4 introduces the Multi-Objective Artificial Bee Colony algorithm. In Section 5 Multi-Objective Gravitational Search Algorithm is explained. Section 6 presents the standard Non-dominated Sorting Genetic Algorithm II. Then, several experiments are provided and analyzed in detail in Section 7. Finally the last section summarizes the main conclusions of this work.

Section snippets

Related work

Real important meta-schedulers have been considered to evaluate the goodness of the proposals to fulfill the cost and time requirements demanded by the users. The first one is the Workload Management System (WMS)⁵ due to it belongs to the most extended middleware in Europe gLite – Lightweight Middleware for Grid Computing.⁶ The second meta-scheduler selected for this study is the Deadline Budget Constraint Algorithm (DBC) [4] from Nimrod-G.

Problem statement

Grid environments allow to solve computing problems using several heterogeneous resources coordinated in a decentralized way. Job scheduling is a critical problem in these environments. A proper scheduling algorithm can reduce the response time and the execution cost. Users generally have deadlines and budgets associated to their experiments, but these requirements are in conflict each other. Faster resources usually are more expensive than slower ones. To tackle this problem, a multi-objective

Multi-Objective Gravitational Search Algorithm (MOGSA)

Gravitational Search Algorithm [3] (GSA) is a swarm algorithm from the physical field. Its agents represent planets that have masses with different sizes exerting gravitational attractions among them through different dimensions. These attractions follow the Newtonian gravity law as a metaheuristics. Thus, biggest masses exert more force of attraction than others, positioning themselves as best solutions. In this paper, a multi-objective version, called Multi-Objective Gravitational Search

Multi-Objective Artificial Bee Colony (MOABC)

Artificial Bee Colony (ABC) [1], [2] is a mono-objective swarm algorithm from the biological field. This algorithm is based on the collective behaviour of its agents – bees – to find the best nectar from the flowers. The main feature of the ABC algorithm is that its agents have different behaviours. Some bees move in a multidimensional search space by selecting nectar source considering their last experience and the experience of their hive fellows. However, other bees move randomly without

Non-dominated Sorting Genetic Algorithm (NSGA II)

Non-dominated Sorting Genetic Algorithm II (NSGA II) [4] is the most popular multi-objective genetic algorithm and it is widely known due to its efficiency. In this research, this algorithm is applied to the job scheduling problem to prove the multi-objective goodness of the two proposed swarm algorithms. As a genetic algorithm, it has agents as individuals that compose the evolutionary population. The main steps of NSGA II are shown in Algorithm 3. The NSGA II input has three parameters.

Experiments and results

In this section, results from several experiments are described. The intention of this analysis is the comparison of two swarm algorithms, MOGSA and MOABC, from different fields – physical and biological – with the standard and well-known multi-objective algorithm NSGA II to prove the multi-objective efficiency of these new proposed multi-objective algorithms. These evaluations use six different workflows over two complete and real topologies. Moreover, MOABC, our best algorithm, is compared

Conclusions and future work

This paper studies and compares three meta-scheduler algorithms based on multi-objective approaches. Two algorithms are inspired from the biology and physics fields – MOABC and MOGSA – that work according to swarm behaviour. Also a popular multi-objective genetic algorithm – NSGA II – is compared with the mentioned algorithms to evaluate the goodness of them. MOABC highlights because of its set coverage and hypervolume, being superior in all the cases than the other multi-objective algorithms.

References (26)

E. Rashedi et al.
GSA: a gravitational search algorithm
Information Sciences
(2009)
M. Wieczorek et al.
Towards a general model of the multi-criteria workflow scheduling on the grid
Future Generation Computer Systems
(2009)
D. Abramson et al.
A computational economy for grid computing and its implementation in the Nimrod-G resource broker
Future Generation Computer Systems
(2002)
F. Xhafa et al.
Computational models and heuristic methods for grid scheduling problems
Future Generation Computer Systems
(2010)
A. Zhou et al.
Multiobjective evolutionary algorithms: a survey of the state of the art
Swarm and Evolutionary Computation
(2011)
D. Kundu et al.
Multi-objective optimization with artificial weed colonies
Information Sciences
(2011)
P. Chakraborty et al.
On convergence of the multi-objective particle swarm optimizer
Information Sciences
(2011)
T. Tsuchiya et al.
Genetics-based multiprocessor scheduling using task duplication
Microprocessors and Microsystems
(1998)
A. Sulistio et al.
On incorporating differentiated levels of network service into gridsim
Future Generation Computer Systems
(2007)
D. Karaboga et al.
A powerful and efficient algorithm for numerical function optimization: Artificial Bee Colony (ABC) algorithm
Journal of Global Optimization
(2007)

D. Karaboga et al.

Artificial Bee Colony (ABC) optimization algorithm for solving constrained optimization problems

K. Deb et al.

A fast elitist multi-objective genetic algorithm: NSGA-II

IEEE Transactions on Evolutionary Computation

(2000)

R. Buyya et al.

GridSim: a toolkit for the modeling and simulation of distributed resource management and scheduling for grid computing

Concurrency and Computation: Practice and Experience

(2002)

Cited by (13)

A framework for multi-objective optimisation based on a new self-adaptive particle swarm optimisation algorithm
2017, Information Sciences
Citation Excerpt :
Over the last few decades, the multiple-objective optimisation (MOO) has gained great attentions in different areas such as manufacturing optimisation [1,2] and environmental/economic dispatch [3].
This paper develops a particle swarm optimisation (PSO) based framework for multi-objective optimisation (MOO). As a part of development, a new PSO method, named self-adaptive PSO (SAPSO), is first proposed. Since the convergence of SAPSO determines the quality of the obtained Pareto front, this paper analytically investigates the convergence of SAPSO and provides a parameter selection principle that guarantees the convergence. Leveraging the proposed SAPSO, this paper then designs a SAPSO-based MOO framework, named SAMOPSO. To gain a well-distributed Pareto front, we also design an external repository that keeps the non-dominated solutions. Next, a circular sorting method, which is integrated with the elitist-preserving approach, is designed to update the external repository in the developed MOO framework. The performance of the SAMOPSO framework is validated through 12 benchmark test functions and a real-word MOO problem. For rigorous validation, the performance of the proposed framework is compared with those of four well-known MOO algorithms. The simulation results confirm that the proposed SAMOPSO outperforms its contenders with respect to the quality of the Pareto front over the majority of the studied cases. The non-parametric comparison results reveal that the proposed method is significantly better than the four algorithms compared at the confidence level of 90% over the 12 test functions.
Bi-objective workflow scheduling of the energy consumption and reliability in heterogeneous computing systems
2017, Information Sciences
Citation Excerpt :
For scheduling tasks with precedence constraints, a general Pareto front approximation was devised. On the basis of multi-objective swarm algorithms, Arsuaga-Ríos et al. [3] presented the multi-objective artificial bee colony and the multi-objective gravitational search algorithm in search of a bi-objective optimization for both execution time and cost. Durillo et al. [10] proposed a novel multi-objective heterogeneous earliest finish time (MOHEFT) list-based workflow scheduling algorithm, which is an extension of the heterogeneous earliest finish time (HEFT) [26] algorithm.
Recent studies focus primarily on low energy consumption or execution time for task scheduling with precedence constraints in heterogeneous computing systems. In most cases, system reliability is more important than other performance metrics. In addition, energy consumption and system reliability are two conflicting objectives. A novel bi-objective genetic algorithm (BOGA) to pursue low energy consumption and high system reliability for workflow scheduling is presented in this paper. The proposed BOGA offers users more flexibility when jobs are submitted to a data center. On the basis of real-world and randomly generated application graphs, numerous experiments are conducted to evaluate the performance of the proposed algorithm. In comparison with excellent algorithms such as multi-objective heterogeneous earliest finish time (MOHEFT) and multi-objective differential evolution (MODE), BOGA performs significantly better in terms of finding the spread of compromise solutions.
Elite-guided multi-objective artificial bee colony algorithm
2015, Applied Soft Computing
Citation Excerpt :
Multi-objective optimization, which involves more than one objective function, has been applied in many fields such as manufacturing optimization [1,2], engineering design [3], and chemical engineering [4].
Multi-objective optimization has been a difficult problem and a research focus in the field of science and engineering. This paper presents a novel multi-objective optimization algorithm called elite-guided multi-objective artificial bee colony (EMOABC) algorithm. In our proposal, the fast non-dominated sorting and population selection strategy are applied to measure the quality of the solution and select the better ones. The elite-guided solution generation strategy is designed to exploit the neighborhood of the existing solutions based on the guidance of the elite. Furthermore, a novel fitness calculation method is presented to calculate the selecting probability for onlookers. The proposed algorithm is validated on benchmark functions in terms of four indicators: GD, ER, SPR, and TI. The experimental results show that the proposed approach can find solutions with competitive convergence and diversity within a shorter period of time, compared with the traditional multi-objective algorithms. Consequently, it can be considered as a viable alternative to solve the multi-objective optimization problems.
Convergence analysis and performance of an improved gravitational search algorithm
2014, Applied Soft Computing Journal
Citation Excerpt :
As pointed out above, nature-inspired heuristic algorithms mimic physical, collective and biological phenomenon and seek good results in specific optimization problems at a reasonable computational load through defined iteration equations and specified operators. Some of these famous physics-inspired heuristic algorithms are SA [11], EM [12], CFO [13–17,46], APO [22,23] and GSA [28–44]. Among them, different points of views based on the selected characteristics have been considered to describe these physic-based algorithms as below.
Gravitational search algorithm (GSA) has been shown to yield good performance for solving various optimization problems. However, it tends to suffer from premature convergence and loses the abilities of exploration and exploitation when solving complex problems. This paper presents an improved gravitational search algorithm (IGSA) that first employs chaotic perturbation operator and then considers memory strategy to overcome the aforementioned problems. The chaotic operator can enhance its global convergence to escape from local optima, and the memory strategy provides a faster convergence and shares individual's best fitness history to improve the exploitation ability. After that, convergence analysis of the proposed IGSA is presented based on discrete-time linear system theory and results show that IGSA is not only guaranteed to converge under the conditions, but can converge to the global optima with the probability 1. Finally, choice of reasonable parameters for IGSA is discussed on four typical benchmark test functions based on sensitivity analysis. Moreover, IGSA is tested against a suite of benchmark functions with excellent results and is compared to GA, PSO, HS, WDO, CFO, APO and other well-known GSA variants presented in the literatures. The results obtained show that IGSA converges faster than GSA and other heuristic algorithms investigated in this paper with higher global optimization performance.
Design, Characterization, and Optimization of the Soft Bellows Pneumatic Actuator
2023, 2023 IEEE International Conference on Mechatronics and Automation, ICMA 2023
Combinatorial double auction based meta-scheduler for medical image analysis application in grid environment
2020, Recent Advances in Computer Science and Communications

View all citing articles on Scopus

María Arsuaga-Ríos is Computer Engineer from the University of Murcia (Spain). She got another two MSc. The first one was related to Grid Computing and e-Engineering at Cranfield University, (UK), in 2008. And the second MSc was about Information Technologies and Advanced Telematics at University of Murcia (Spain) in 2009. After her studies, she has been working on projects related to Data Mining, Ontology Systems, Bioinspired Algorithms and Optimization. She is doing her PhD with the University of Extremadura (Spain). Her PhD consists in researching different multi-objective strategies based on bioinspired algorithms to optimize the job scheduling problem in Grid environments. At the same time, she have taught artificial intelligence applied to Grid Computing in a MSc at the University of Extremadura. Currently, she is also enrolled in the project FESA that is being developed at CERN (Switzerland).

Miguel A. Vega-Rodríguez is a professor of Computer Architecture in the Department of Computer and Communications Technologies, University of Extremadura, Spain. He received a PhD degree in Computer Science from the University of Extremadura. Dr. Vega-Rodríguez has authored or co-authored more than 420 publications including journal papers, book chapters and peer-reviewed conference proceedings. In addition, he is editor and reviewer of several international JCR journals. Dr. Vega-Rodríguez's main research interests are parallel and distributed computing, reconfigurable computing, and evolutionary computing.

Francisco Prieto-Castrillo studied theoretical physics at Universidad Autónoma de Madrid. He obtained his Ph.D. in physics of complex systems and nonlinear dynamics. He has been involved for 6 years in the study of complex systems and their applications through different disciplines ranging from earthquake engineering to grid computing. At the present time his research is focussed on the optimization of distributed computing networks through complex systems based techniques.

¹: Tel.: +34 927 25 72 63; fax: +34 927 25 71 87.

²: Tel.: +34 927 65 93 17; fax: +34 927 32 32 37.

View full text

Meta-schedulers for grid computing based on multi-objective swarm algorithms

Abstract

Graphical abstract

Highlights

Introduction

Section snippets

Related work

Problem statement

Multi-Objective Gravitational Search Algorithm (MOGSA)

Multi-Objective Artificial Bee Colony (MOABC)

Non-dominated Sorting Genetic Algorithm (NSGA II)

Experiments and results

Conclusions and future work

Information Sciences

Future Generation Computer Systems

Future Generation Computer Systems

Future Generation Computer Systems

Swarm and Evolutionary Computation

Information Sciences

Information Sciences

Microprocessors and Microsystems

Future Generation Computer Systems

A powerful and efficient algorithm for numerical function optimization: Artificial Bee Colony (ABC) algorithm

Journal of Global Optimization

Artificial Bee Colony (ABC) optimization algorithm for solving constrained optimization problems

A fast elitist multi-objective genetic algorithm: NSGA-II

IEEE Transactions on Evolutionary Computation

GridSim: a toolkit for the modeling and simulation of distributed resource management and scheduling for grid computing

Concurrency and Computation: Practice and Experience