Abstract. Data centres are facilities with large amount of machines (i.e., servers) and hosted processes (e.g., virtual machines). Managers of data centres (e.g., operators, capital allocators, CRM) constantly try to optimise them, reassigning 'better' machines to processes. These managers usually see better/good placements as a combination of distinct objectives, hence why in this paper we define the data centre optimisation problem as a multi-objective machine reassignment problem. While classical solutions to address this either do not find many solutions (e.g., GRASP), do not cover well the search space (e.g., PLS), or even cannot operate properly (e.g., NSGA-II lacks a good initial population), we propose GeNePi, a novel hybrid algorithm. We show that GeNePi outperforms all the other algorithms in terms of quantity of solutions (nearly 6 times more solutions on average than the second best algorithm) and quality (hypervolume of the Pareto frontier is 106% better on average).

Optimising the data centres of large IT organisations is complex as (i) they are composed of various hosting departments with their own preferences and (ii) reassignment solutions can be evaluated from various independent dimensions. But in reality, the problem is even more challenging as companies can now choose from a pool of cloud services to host some of their workloads. This hybrid search space seems intractable, as each workload placement decision (seen as running in a virtual machine on a server) is required to answer many questions: can we host it internally? In which hosting department? Are the capital allocators of this hosting department ok with this placement? How much does it save us and is it safe? Is there a better option in the Cloud? Etc. In this paper, we define the multi-objective VM reassignment problem for hybrid and decentralised data centres. We also propose H2-D2, a solution that uses a multi-layer architecture and a metaheuristic algorithm to suggest reassignment solutions that are evaluated by the various hosting departments (according to their preferences). We compare H2-D2 against state-of-the-art multi-objective algorithms and find that H2-D2 outperforms them both in terms of quantity (approx 30% more than the second-best algorithm on average) and quality of solutions (19% better than the second-best on average).

Context: When software architects or engineers are given a list of all the features and their interactions (i.e., a Feature Model or FM) together with stakeholders' preferences-their task is to find a set of potential products to suggest the decision makers. Software Product Lines Engineering (SPLE) consists in optimising those large and highly constrained search spaces according to multiple objectives reflecting the preference of the different stakeholders. SPLE is known to be extremely skill-and labour-intensive and it has been a popular topic of research in the past years. Objective: This paper presents the first thorough description and evaluation of the related problem of evolving software product lines. While change and evolution of software systems is the common case in the industry, to the best of our knowledge this element has been overlooked in the literature. In particular, we evaluate whether seeding previous solutions to genetic algorithms (that work well on the general problem) would help them to find better/faster solutions. Method: We describe in this paper a benchmark of large scale evolving FMs, consisting of 5 popular FMs and their evolutions-synthetically generated following an experimental study of FM evolution. We then study the performance of a state-of-the-art algorithm for multi-objective FM selection (SATIBEA) when seeded with former solutions. Results: Our experiments show that we can improve both the execution time and the quality of SATIBEA by feeding it with previous configurations. In particular, SATIBEA with seeds proves to converge an order of magnitude faster than SATIBEA alone. Conclusion: We show in this paper that evolution of FMs is not a trivial task and that seeding previous solutions can be used as a first step in the optimisation-unless the difference between former and current FMs is high, where seeding has a limited impact.

Testing is crucial to ensure the quality of software systems -but testing is an expensive process, so test managers try to minimise the set of tests to run to save computing resources and speed up the testing process and analysis. One problem is that there are different perspectives on what is a good test and it is usually not possible to compare these dimensions. This is a perfect example of a multi-objective optimisation problem, which is hard -especially given the scale of the search space here. In this paper, we propose a novel hybrid algorithm to address this problem. Our method is composed of three steps: a greedy algorithm to find quickly some good solutions, a genetic algorithm to increase the search space covered and a local search algorithm to refine the solutions. We demonstrate through a large scale empirical evaluation that our method is more reliable (better whatever the time budget) and more robust (better whatever the number of dimensions considered) -in the scenario with 4 objectives and a default execution time, we are 178% better in hypervolume on average than the state-of-the-art algorithms.

Machine Reassignment is a challenging problem for constraint programming (CP) and mixed integer linear programming (MILP) approaches, especially given the size of data centres. The multi-objective version of the Machine Reassignment Problem is even more challenging and it seems unlikely for CP or MILP to obtain good results in this context. As a result, the first approaches to address this problem have been based on other optimisation methods, including metaheuristics. In this paper we study under which conditions a mixed integer optimisation solver, such as IBM ILOG CPLEX, can be used for the Multi-objective Machine Reassignment Problem. We show that it is useful only for small or medium scale data centres and with some relaxations, such as an optimality tolerance gap and a limited number of directions explored in the search space. Building on this study, we also investigate a hybrid approach, feeding a metaheuristic with the results of CPLEX, and we show that the gains are important in terms of quality of the set of Pareto solutions (+126.9% against the metaheuristic alone and +17.8% against CPLEX alone) and number of solutions (8.9 times more than CPLEX), while the processing time increases only by 6% in comparison to CPLEX for execution times larger than 100 seconds.