Swarm Intelligence And Its Applications In Distributed Systems

Swarm intelligence computing as a new method of solving problems, has anirreplaceable role in the field of engineering optimization. It is very important in solvinglarge-scale optimization problem because of improved the convergence speed andoptimized the accuracy of result set. However, the complexity of the specificoptimization problem always either makes the algorithm convergence speed get ratherslow, or leads resultant to marginal precision. Therefore, when encountering suchpractical problems, the researchers need to analyze the characteristics of the problem,use different methods to find answers. Thus, in order to improve the overallperformance of the algorithm for one specific problem, improvement in the operator ofswarm intelligence algorithm and integration of other fusion algorithms are needed.In this paper, we are engaged in discussing meet user QoS constraints in jobsscheduling problem for grid computing and cloud computing environments, andmulti-constrained multicast routing optimization problem. First, we analyzed theshortcoming of two classical algorithms of Swarm Intelligence, the ant colonyalgorithm and particle swarm algorithm. Then, to deal with Jobs scheduling problem ingrid and multicast routing optimization problem, we improved the operators of thesetwo algorithms, integrating some smart methods such as uniform design, divide andconquer algorithm, to speed up the convergence of those algorithms and finallyimproved the overall performance of the application system. The research and the maincontribution are summarized as follows:1) A modified Pareto Ant Colony Optimization (MPACO), based on Pareto AntColony and fuzzy mathematics, was proposed to solve the problem of multicast routingwith QoS constrained in this paper. After analyzing the mathematical model ofinaccurate information network, we used multi-group of ant agents to find thenon-dominated Pareto set which had the maximum probability of meeting QoSrequirments, then randomly selected the best multicast tree from that set. During thecourse of execution, MPACO updated the pheromone values using a new strategy whichwas improved to be more suitable for inaccurate information network. Then at the stage of global pheromone updates, a new incentive-update mechanism for Pareto results wasbeen proposed, in order to improve the convergence speed of algorithm. Simulationresults have shown that the proposed algorithm is both feasible and effective.2) This paper presents an improved Max-Min ant system (IMMAS) to solve thegrid workflow scheduling optimization with various QoS requirements. To improve theMMAS, this paper did not use the traditional method to get feasible schedules, butemploys a divide and conquer method, which redraw sub-deadline of every sub-tasks, tosatisfy the customer's deadline and cost requirements. Besides, a new heuritic updaterule is embedded to improve the convergence of algorithm. And the experimentsconfirmed the effectiveness of the proposed algorithm.3) One of the key technologies to improve the efficiency of Grid computing is tosolve the independent Job scheduling problem under the QoS constraints. Therefore, inthis paper we designed a novel algorithm named UDPSO(Uniform-Design DiscreteParticle Swarm Optimization)to find a sufficient number of uniformly distributed andrepresentative Pareto optimal solution for the problem. Using that Pareto set, theperformance of gird system will be more efficient. To solve the problem, we redefinedthe velocity and position of particles, and used a new method to optimize those twoparameters. Then, employ uniform design method to get distributed Pareto front inobjective space efficiently, and finally the global convergence of the algorithm has beenproved. Simulation results showed that this algorithm is effective and practical.4) Proposed an Improved Discrete Particle of Swarm Optimization (IDPSO)algorithm to optimize the job scheduling problem in cloud computing with user prioritylevel preferences. The user priority and the task deadline were combined to establish anappropriate task priority to guide the algorithm fitness function. Then employed are-optimization criterion to ensure that the algorithm has the ability to jump out of localoptima, and ultimately obtained task scheduling mapping with user-priority preference.Simulation results show that this algorithm is effective and practical.5) Analyze the complexity of Continuous Range Neighbor query processingstrategies in the Time-Square Distance space. We conduct the analysis with twodifferent mobility models, which are the Linear model and the Piecewise-Linear model,respectively. For each model we consider two query types, namely On-line query andOff-line query. For each model and each query type, we analyze the complexity of initial construction and that of updates. Then based on the result we obtain in this paper,we propose two efficient algorithms to process range neighbor queries, which havecomplexity of O (Nlog(N))and O (m2Nlog(mN)), respectively.