Now showing 1 - 9 of 9
  • Publication
    Self-Balancing Decentralized Distributed Platform for Urban Traffic Simulation
    Microscopic traffic simulation is the most accurate tool for predictive analytics in urban environments. However, the amount of workload (i.e., cars simulated simultaneously) can be challenging for classical systems, particularly for scenarios requiring faster than real-time processing (e.g., for emergency units having to make quick decisions on traffic management). This challenge can be tackled with distributed simulations by sharing the load between simulation engines running on different computing nodes, hence balancing the processing power required. This paper studies the performance of dSUMO, i.e., a distributed microscopic traffic simulator. dSUMO is fully decentralized and can dynamically balance the workload between its computing nodes, hence showing important improvements against classical, centralized and not dynamic, solutions.
    Scopus© Citations 14  692
  • Publication
    Towards a Multi-Objective VM Reassignment for Large Decentralised Data Centres
    Optimising the IT infrastructure of large, often geographically distributed, organisations goes beyond the classical virtual machine reassignment problem, for two reasons: (i) the data centres of these organisations are composed of a number of hosting departments which have different preferences on what to host and where to host it; (ii) the top-level managers in these data centres make complex decisions and need to manipulate possible solutions favouring different objectives to find the right balance. This challenge has not yet been comprehensively addressed in the literature and in this paper we demonstrate that a multi-objective VM reassignment is feasible for large decentralised data centres. We show on a realistic data set that our solution outperforms other classical multi-objective algorithms for VM reassignment in terms of quantity of solutions (by about 15% on average) and quality of the solutions set (by over 6% on average). 
    Scopus© Citations 10  585
  • Publication
    A comparative study of multi-objective machine reassignment algorithms for data centres
    At a high level, data centres are large IT facilities hosting physical machines (servers) that often run a large number of virtual machines (VMs)— but at a lower level, data centres are an intricate collection of interconnected and virtualised computers, connected services, complex service-level agreements. While data centre managers know that reassigning VMs to the servers that would best serve them and also minimise some cost for the company can potentially save a lot of money—the search space is large and constrained, and the decision complicated as they involve different dimensions. This paper consists of a comparative study of heuristics and exact algorithms for the Multi-objective Machine Reassignment problem. Given the common intuition that the problem is too complicated for exact resolutions, all previous works have focused on various (meta)heuristics such as First-Fit, GRASP, NSGA-II or PLS. In this paper, we show that the state-of-art solution to the single objective formulation of the problem (CBLNS) and the classical multi-objective solutions fail to bridge the gap between the number, quality and variety of solutions. Hybrid metaheuristics, on the other hand, have proven to be more effective and efficient to address the problem – but as there has never been any study of an exact resolution, it was difficult to qualify their results. In this paper, we present the most relevant techniques used to address the problem, and we compare them to an exact resolution ( -Constraints). We show that the problem is indeed large and constrained (we ran our algorithm for 30 days on a powerful node of a supercomputer and did not get the final solution for most instances of our problem) but that a metaheuristic (GeNePi) obtains acceptable results: more (+188%) solutions than the exact resolution and a little more than half (52%) the hypervolume (measure of quality of the solution set).
      476Scopus© Citations 6
  • Publication
    Towards the Automatic Detection of Efficient Computing Assets in a Heterogeneous Cloud Environment
    (Institute of Electrical and Electronic Engineers (IEEE), 2013-06-03) ; ; ; ;
    In a heterogeneous cloud environment, the manual grading of computing assets is the first step in the process of configuring IT infrastructures to ensure optimal utilization of resources. Grading the efficiency of computing assets is however, a difficult, subjective and time consuming manual task. Thus, an automatic efficiency grading algorithm is highly desirable. In this paper, we compare the effectiveness of the different criteria used in the manual grading task for automatically determining the efficiency grading of a computing asset. We report results on a dataset of 1,200 assets from two different data centers in IBM Toronto. Our preliminary results show that electrical costs (associated with power and cooling) appear to be even more informative than hardware and age based criteria as a means of determining the efficiency grade of an asset. Our analysis also indicates that the effectiveness of the various efficiency criteria is dependent on the asset demographic of the data centre under consideration.
      343
  • Publication
    Global dynamic load-balancing for decentralised distributed simulation
    (Institute of Electrical and Electronic Engineers (IEEE), 2014-12-10) ; ;
    Distributed simulations require partitioning mechanisms to operate, and the best partitioning algorithms try to load-balance the partitions. Dynamic load-balancing, i.e. re-partitioning simulation environments at run-time, becomes essential when the load in the partitions change. In decentralised distributed simulation the information needed to dynamically load-balance seems difficult to collect and to our knowledge, all solutions apply a local dynamic load balancing: partitions exchange load only with their neighbours (more loaded partitions to less loaded ones). This limits the effect of the load-balancing. In this paper, we present a global dynamic load-balancing of decentralised distributed simulations. Our algorithm collects information in a decentralised fashion and makes re-balancing decisions based on the load processed by every logical processes. While our algorithm has similar results to others in most cases, we show an improvement of the load-balancing up to 30% in some challenging scenarios against only 12.5% for a local dynamic load-balancing.
    Scopus© Citations 3  369
  • Publication
    VM reassignment in hybrid clouds for large decentralised companies: A multi-objective challenge
    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).
      555Scopus© Citations 31
  • Publication
    Synchronisation for Dynamic Load Balancing of Decentralised Conservative Distributed Simulation
    (Association for Computing Machinery, 2014-05-21) ; ;
    Synchronisation mechanisms are essential in distributed simulation. Some systems rely on central units to control the simulation but central units are known to be bottlenecks [10]. If we want to avoid using a central unit to optimise the simulation speed, we lose the capacity to act on the simulation at a global scale. Being able to act on the entire simulation is an important feature which allows to dynamically load-balance a distributed simulation. While some local partitioning algorithms exist [12], their lack of global view reduces their efficiency. Running a global partitioning algorithm without central unit requires a synchronisation of all logical processes (LPs) at the same step.We introduce in this paper two algorithms allowing to synchronise logical processes in a distributed simulation without any central unit. The first algorithm requires the knowledge of some topological properties of the network while the second algorithm works without any requirement. The algorithms are detailed and compared against each other. An evaluation shows the benefits of using a global dynamic load-balancing for distributed simulations.
      347Scopus© Citations 5
  • Publication
    MILP for the Multi-objective VM Reassignment Problem
    Machine Reassignment is a challenging problem for constraint programming (CP) and mixed integer linear pro- gramming (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.
    Scopus© Citations 13  633
  • Publication
    dSUMO: Towards a Distributed SUMO
    Microscopic urban mobility simulations consist of modelling a city's road network and infrastructure, and to run autonomous individual vehicles to understand accurately what is going on in the city. However, when the scale of the problem space is large or when the processing time is critical, performing such simulations might be problematic as they are very computationally expensive applications. In this paper, we propose to leverage the power of many computing resources to perform quicker or larger microscopic simulations, keeping the same accuracy as the classical simulation running on a single computing unit. We have implemented a distributed version of SUMO, called dSUMO. We show in this paper that the accuracy of the simulation in SUMO is not impacted by the distribution and we give some preliminary results regarding the performance of dSUMO compared to SUMO.
      727