Minimizing the resource wastage reduces the energy cost of operating a data center, but may also lead to a considerably high resource overcommitment affecting the Quality of Service (QoS) of the running applications. The effective tradeoff between resource wastage and overcommitment is a challenging task in virtualized Clouds and depends on the allocation of virtual machines (VMs) to physical resources. We propose in this paper a multi-objective method for dynamic VM placement, which exploits live migration mechanisms to simultaneously optimize the resource wastage, overcommitment ratio and migration energy. Our optimization algorithm uses a novel evolutionary meta-heuristic based on an island population model to approximate the Pareto optimal set of VM placements with good accuracy and diversity. Simulation results using traces collected from a real Google cluster demonstrate that our method outperforms related approaches by reducing the migration energy by up to 57% with a QoS increase below 6%.

Many techniques such as scheduling and resource provisioning rely on performance prediction of workflow tasks for varying input data. However, such estimates are difficult to generate in the cloud. This paper introduces a novel two-stage machine learning approach for predicting workflow task execution times for varying input data in the cloud. In order to achieve high accuracy predictions, our approach relies on parameters reflecting runtime information and two stages of predictions. Empirical results for four real world workflow applications and several commercial cloud providers demonstrate that our approach outperforms existing prediction methods. In our experiments, our approach respectively achieves a best-case and worst-case estimation error of 1.6 and 12.2 percent, while existing methods achieved errors beyond 20 percent (for some cases even over 50 percent) in more than 75 percent of the evaluated workflow tasks. In addition, we show that the models predicted by our approach for a specific cloud can be ported with low effort to new clouds with low errors by requiring only a small number of executions.

Minimizing the resource wastage reduces the energy cost of operating a data center, but may also lead to a considerably high resource overcommitment affecting the Quality of Service (QoS) of the running applications. Determining the effective tradeoff between resource wastage and overcommitment is a challenging task in virtualized Cloud data centers and depends on how Virtual Machines (VMs) are allocated to physical resources. In this paper, we propose a multi-objective framework for dynamic placement of VMs exploiting live-migration mechanisms which simultaneously optimize the resource wastage, overcommitment ratio and migration cost. The optimization algorithm is based on a novel evolutionary meta-heuristic using an island population model underneath. We implemented and validated our method based on an enhanced version of a well-known simulator. The results demonstrate that our approach outperforms other related approaches by reducing up to 57% migrations energy consumption while achieving different energy and QoS goals.

Stencil computations expose a large and complex space of equivalent implementations. These computations often rely on autotuning techniques, based on iterative compilation or machine learning (ML), to achieve high performance. Iterative compilation autotuning is a challenging and time-consuming task that may be unaffordable in many scenarios. Meanwhile, traditional ML autotuning approaches exploiting classification algorithms (such as neural networks and support vector machines) face difficulties in capturing all features of large search spaces. This paper proposes a new way of automatically tuning stencil computations based on structural learning. By organizing the training data in a set of partially-sorted samples (i.e., rankings), the problem is formulated as a ranking prediction model, which translates to an ordinal regression problem. Our approach can be coupled with an iterative compilation method or used as a standalone autotuner. We demonstrate its potential by comparing it with state-of-the-art iterative compilation methods on a set of nine stencil codes and by analyzing the quality of the obtained ranking in terms of Kendall rank correlation coefficients.

The increasing performance of today's computer architecture comes with an unprecedented augment of hardware complexity. Unfortunately this results in difficult-to-tune software and consequentially in a gap between the potential peak performance and the actual performance. Automatic tuning is an emerging approach that assists the programmer in managing this complexity. State-of-the-art autotuners are limited, though: they either require long tuning times, e.g., due to iterative searches, or cannot tackle the complexity of the problem due to the limitation of the supervised machine learning (ML) methodologies used. In particular, traditional ML autotuning approaches exploiting classification algorithms (such as neural networks and support vector machines) face difficulties in capturing all features of large search spaces. We propose a new way of performing automatic tuning based on structural learning: the tuning problem is formulated as a version ranking prediction modeling and solved using ordinal regression. We demonstrate its potential on a well-known autotuning problem: stencil computations. We compare state-of-the-art iterative compilation methods with our ordinal regression approach and analyze the quality of the obtained ranking in terms of Kendall rank correlation coefficients.

Multiple sequence alignment (MSA) plays a core role in most bioinformatics studies and provides a framework for the analysis of evolution in biological systems. The MSA problem consists in finding an optimal alignment of three or more sequences of nucleotides or amino acids. Different scores have been defined to assess the quality of MSA solutions, so the problem can be formulated as a multiobjective optimization problem. The number of proposals focused on this approach in the literature is scarce, and most of the works take as base algorithm the NSGA-II metaheuristic. So, there is a lack of a study involving a set of representative multiobjective metaheuristics to deal with this complex problem. Our main goal in this paper is to carry out such study. We propose a biobjective formulation for the MSA and perform an exhaustive comparative study of six multiobjective algorithms. We have considered a number of problems taken from the benchmark BAliBASE (v3.0). Our experiments reveal that the classic NSGA-II algorithm and MOCell, a cellular metaheuristic, provide the best overall performance.

The design of bar structures in civil engineering is a complex problem when dealing with real-world structures. An approach to deal with these problems is to apply metaheuristics, which are stochastic methods based on iteratively producing and evaluating tentative solutions. In particular, we focus on multi-objective metaheuristics, as we consider two goals to be minimized: the weight and the deflection of the structure. When applying these techniques to real-world problems, running a metaheuristic for several thousands of evaluations may require many days on a single processor. In this paper, we develop distributed master/slave versions of four multi-objective metaheuristics that are representative of the state-of-the-art and apply them to optimize the design of two instances of a cable-strayed bridge. Our study reveals that our parallel proposals are able to effectively use up to 450 cores, providing accurate solutions in a short time.

While in the past scheduling algorithms were almost exclusively targeted at optimizing applications' make span, today they must simultaneously optimise several goals. Among these goals, energy efficiency is receiving increasing attention for environmental and financial reasons. In contrast to related work that optimises energy consumption as a single objective function, we reformulate in this paper the problem as a bi-objective optimisation by considering both make span and energy as goals. We study the potential benefits of using a Pareto-based workflow scheduling algorithm called MOHEFT using realistic energy consumption and performance models for task executions. We analyse the tradeoff solutions computed by MOHET for different workflows (different in shapes and sizes) in different execution scenarios (different resources in terms of energy consumption). The obtained results show that our bi-objective approach found in some cases schedules that reduce the energy consumption up to 85% with only 3.3% of make span concessions.

Algorithms based on the Particle Swarm Optimization (PSO) scheme have become popular to solve both single- and multi-objective optimization problems. In this paper, we focus on SMPSO, a PSO designed to cope with this second group of problems. Taking it as our starting point, we analyze different leader selection schemes, which give rise to four new variants of SMPSO. These new versions, along with the original algorithm, are compared using a benchmark composed of 21 problems. Our study reveals that SMPSOhv, a variant that uses the hypervolume indicator to guide leader selection, is the best performing algorithm in our comparison, outperforming also the original version of SMPSO. To further assess the performance of SMPSOhv, we compare it against NSGA-II and SMS-EMOA, achieving again the best overall results in this new comparative study. Based on these observations, we conclude that the use of the hypervolume for leader selection is a promising approach for multi-objective PSO algorithms.

Traditional scheduling research usually targets make span as the only optimization goal, while several isolated efforts addressed the problem by considering at most two objectives. In this paper we propose a general framework and heuristic algorithm for multi-objective static scheduling of scientific workflows in heterogeneous computing environments. The algorithm uses constraints specified by the user for each objective and approximates the optimal solution by applying a double strategy: maximizing the distance to the constraint vector for dominant solutions and minimizing it otherwise. We analyze and classify different objectives with respect to their impact on the optimization process and present a four-objective case study comprising make span, economic cost, energy consumption, and reliability. We implemented the algorithm as part of the ASKALON environment for Grid and Cloud computing. Results for two real-world applications demonstrate that the solutions generated by our algorithm are superior to user-defined constraints most of the time. Moreover, the algorithm outperforms a related bi-criteria heuristic and a bi-criteria genetic algorithm.

Nowadays, scientists and companies are confronted with multiple competing goals such as makespan in high-performance computing and economic cost in Clouds that have to be simultaneously optimized. Multi-objective scheduling of scientific workflows in distributed systems is therefore receiving increasing research attention. Most existing approaches typically aggregate all objectives in a single function, defined a-priori without any knowledge about the problem being solved, which negatively impacts the quality of the solutions. In contrast, Pareto-based approaches having as outcome a set of several (nearly-) optimal solutions that represent a tradeoff among the different objectives, have been scarcely studied. In this paper, we propose a new Pareto-based list scheduling heuristic that provides the user with a set of tradeoff optimal solutions from where the one that better suits the user requirements can be manually selected. We demonstrate the potential of MOHEFT for a bi-objective scheduling problem that optimizes makespan and economic cost in a Cloud-based computing scenario. We compare MOHEFT with two state-of-the-art approaches using different synthetic and real-world workflows: the classical HEFT algorithm used in single-objective scheduling and the SPEA2* genetic algorithm used for multi-objective optimisation problems.

In this paper we introduce a multi-objective autotuning framework comprising compiler and runtime components. Focusing on individual code regions, our compiler uses a novel search technique to compute a set of optimal solutions, which are encoded into a multi-versioned executable. This enables the runtime system to choose specifically tuned code versions when dynamically adjusting to changing circumstances. We demonstrate our method by tuning loop tiling in cache-sensitive parallel programs, optimizing for both runtime and efficiency. Our static optimizer finds solutions matching or surpassing those determined by exhaustively sampling the search space on a regular grid, while using less than 4% of the computational effort on average. Additionally, we show that parallelism-aware multi-versioning approaches like our own gain a performance improvement of up to 70% over solutions tuned for only one specific number of threads.

jMetal is a Java-based framework for multi-objective optimization using metaheuristics. It is a flexible, extensible, and easy-to-use software package that has been used in a wide range of applications. In this paper, we describe the design issues underlying jMetal, focusing mainly on its internal architecture, with the aim of offering a comprehensive view of its main features to interested researchers. Among the covered topics, we detail the basic components facilitating the implementation of multi-objective metaheuristics (solution representations, operators, problems, density estimators, archives), the included quality indicators to assess the performance of the algorithms, and jMetal's support to carry out full experimental studies.

SBSE techniques have been widely applied to requirements selection and prioritization problems in order to ascertain a suitable set of requirements for the next release of a system. Unfortunately, it has been widely observed that requirements tend to be changed as the development process proceeds and what is suitable for today, may not serve well into the future. Though SBSE has been widely applied to requirements analysis, there has been no previous work that seeks to balance the requirements needs of today with those of the future. This paper addresses this problem. It introduces a multi-objective formulation of the problem which is implemented using multi-objective Pareto optimal evolutionary algorithms. The paper presents the results of experiments on both synthetic and real world data. Copyright 2010 ACM.

To evaluate the search capabilities of a multiobjective algorithm, the usual approach is to choose a benchmark of known problems, to perform a fixed number of function evaluations, and to apply a set of quality indicators. However, while real problems could have hundreds or even thousands of decision variables, current benchmarks are normally adopted with relatively few decision variables (normally from 10 to 30). Furthermore, performing a constant number of evaluations does not provide information about the effort required by an algorithm to get a satisfactory set of solutions; this information would also be of interest in real scenarios, where evaluating the functions defining the problem can be computationally expensive. In this paper, we study the effect of parameter scalability in a number of state-of-the-art multiobjective metaheuristics. We adopt a benchmark of parameter-wise scalable problems (the Zitzler-Deb-Thiele test suite) and analyze the behavior of eight multiobjective metaheuristics on these test problems when using a number of decision variables that range from 8 up to 2048. By using the hypervolume indicator as a stopping condition, we also analyze the computational effort required by each algorithm in order to reach the Pareto front. We conclude that the two analyzed algorithms based on particle swarm optimization and differential evolution yield the best overall results.

This paper introduces a new cellular genetic algorithm for solving multiobjective continuous optimization problems. Our approach is characterized by using an external archive to store nondominated solutions and a feedback mechanism in which solutions from this archive randomly replace existing individuals in the population after each iteration. The result is a simple and elitist algorithm called MOCell. Our proposal has been evaluated with both constrained and unconstrained problems and compared against NSGA-II and SPEA2, two state-of-the-art evolutionary multiobjective optimizers. For the studied benchmark, our experiments indicate that MOCell obtains competitive results in terms of convergence and hypervolume, and it clearly outperforms the other two compared algorithms concerning the diversity of the solutions along the Pareto front. © 2009 Wiley Periodicals, Inc.

Some real-world optimization problems have hundreds or even thousands of decision variables. However, the effect that the scalability of parameters has in modern multi-objective metaheuristic algorithms has not been properly studied (the current benchmarks are normally adopted with ten to thirty decision variables). In this paper, we adopt a benchmark of parameter-wise scalable problems (the ZDT test problems) and analyze the behavior of six multi-objective metaheuristics on these test problems when using a number of decision variables that goes from 8 up to 2048. The computational effort required by each algorithm in order to reach the true Pareto front is also analyzed. Our study concludes that a particle swarm algorithm provides the best overall performance, although it has difficulties in multifrontal problems.

Many of the optimization problems from the real world are multiobjective in nature, and the reference algorithm for multiobjective optimization is NSGA-II. Frequently, these problems present a high complexity, so classical metaheuristic algorithms fail to solve them in a reasonable amount of time; in this context, parallelism is a choice to overcome this fact to some extent. In this paper we study three parallel approaches (a synchronous and two asynchronous strategies) for the NSGA-II algorithm based on the master-worker paradigm. The asynchronous schemes are designed to be used in grid systems, so they can make use of hundreds of machines. We have applied them to solve a real world problem which lies in optimizing a broadcasting protocol using a network simulator. Our experiences reveal that significant time reductions can be achieved with the distributed approaches by using a grid system of more than 300 processors.

We propose the use of a new algorithm to solve multiobjective optimization problems. Our proposal adapts the well-known scatter search template for single-objective optimization to the multiobjective domain. The result is a hybrid metaheuristic algorithm called Archive-Based hYbrid Scatter Search (AbYSS), which follows the scatter search structure but uses mutation and crossover operators from evolutionary algorithms. AbYSS incorporates typical concepts from the multiobjective field, such as Pareto dominance, density estimation, and an external archive to store the nondominated solutions. We evaluate AbYSS with a standard benchmark including both unconstrained and constrained problems, and it is compared with two state-of-the-art multiobjective optimizers, NSGA-II and SPEA2. The results obtained indicate that, according to the benchmark and parameter settings used, AbYSS outperforms the other two algorithms as regards the diversity of the solutions, and it obtains very competitive results according to the convergence to the true Pareto fronts and the hypervolume metric.

Radio network design (RND) is a fundamental problem in cellular networks for telecommunications. In these networks, the terrain must be covered by a set of base stations (or antennae), each of which defines a covered area called cell. The problem may be reduced to figure out the optimal placement of antennae out of a list of candidate sites trying to satisfy two objectives: to maximize the area covered by the radio signal and to reduce the number of used antennae. Consequently, RND is a bi-objective optimization problem. Previous works have solved the problem by using single-objective techniques which combine the values of both objectives. The used techniques have allowed to find optimal solutions according to the defined objective, thus yielding a unique solution instead of the set of Pareto optimal solutions. In this paper, we solve the RND problem using a multi-objective version of the algorithm CHC, which is the metaheuristic having reported the best results when solving the single-objective formulation of RND. This new algorithm, called MOCHC, is compared against a binary-coded NSGA-II algorithm and also against the provided results in the literature. Our experiments indicate that MOCHC outperfoms NSGA-II and, more importantly, it is more efficient finding the optimal solutions than single-objectives techniques.