Font Size: a A A

Research On Service Composition And Selection In Cloud Computing

Posted on:2014-01-07Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y LiuFull Text:PDF
GTID:1228330401463106Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
In cloud computing environment, service composition is an efficient implementation of a composite service. Cloud computing also provides luxuriant atomic services for service composition. In service composition, users’requirements are numerous and complicated. Users’requirements include functional and non-functional demands of composite services. It is quite difficult to select proper atomic services and instances in cloud computing platform. Therefore, it’s a challenging research to select service instances considering multiple constraints and multiple objectives even load balance factors. In the meantime, because of high arrival rate of service composition requests in cloud computing, high real-time service selection methods are inevitable.First, this dissertation reviews service composition research status, especially single-objective and multi-objective service selection problems and Particle Swarm Optimization (PSO) based service selection. Then it analyzes advantages and disadvantages of existing schemes. Since service selection methods hardly cope with various topologies of composite services, we propose a topology conversion mechanism which guarantees non-functional parameter equivalence and facilitates service selection. A Ring Particle Swarm Optimization (RPSO) algorithm based on niching technique is adopted to solve multi-constraint single-objective service selection problems considering users’QoS demands and load balance factors. An Accuracy Sub-swarms Particle Swarm Optimization (ASPSO) algorithm is proposed to resist inherent features of PSO (e.g. premature convergence and diversity loss). ASPSO uses a simple clustering mechanism to construct sub-swarms in the feasible solution dense region to enhance service selection accuracy. Finally, we propose a Lightweight Multi-Objective Particle Swarm Optimization (LMOPSO) algorithm using approximate distance and external archive for multi-objective service selection problems. Major creative work of this dissertation is summarized below:1. QoS-equivalence topology conversion methanism. Due to diverse service composition paths (basic topologies include sequence, parallel, selective, loop etc.) in composite services, service selection algorithms hardly find service instances from original service composition paths according to users’ non-functional parameter demands. We propose an equivalent topology conversion mechanism supporting QoS parameters and systems’ load balance factors. This mechanism converts various service composition paths to aggregated paths which service selection can directly deal with. It presents the construction of aggregated service composition paths and calculation of equivalent non-functional parameters.2. Service selection algorithm based on RPSO. Existing service selection algorithms always depend on problem characteristics, e.g. linearity, nonlinearity, constraint and objective features. Existing research neglects concurrent influence of QoS demands and load balance factors. A niching based RPSO algorithm is adopted for service selection. The proposed algorithm constructs niche with each three adjacent particles, and uses joint particle of each two adjacent niche to form a ring topology. This topology enhances PSO’s resistance on sub-optima in service selection problems and improves accuracy of service selection. Simulation results demonstrate the accuracy of proposed algorithm is higher than the standard PSO, especially in the presence of massive service instances. 3. Service selection algorithm based on ASPSO. Inherent features of PSO (e.g. premature convergence and diversity loss) always induce sub-optima in service selection results. We propose an ASPSO service selection algorithm based on serial and parallel niching techniques. ASPSO divides solution space to multiple grid units, and calculates feasible solution number of each grid unit periodically. Then, ASPSO clusters feasible solution dense units to feasible solution dense regions and constructs sub-swarms in these regions. The algorithm isolates sub-swarms and the main-swarm. It searches for optimal solution using global optima ranking method in all swarms. Simulation results demonstrate ASPSO improves the accuracy of PSO. It decreases impact of premature convergence and diversity loss. With acceptable incremental computation complexity, ASPSO is more accurate than compared algorithms in multi-constraint single-objective service selection problems.4. Service selection algorithm based on LMPSO. For users’ multi-objective service composition requests, existing methods mainly convert multi-objective problems to multi-constraint single-objective ones. But these methods take effect with priori knowledge of solution spaces. Each existing method only outputs one solution after each execution. Users hardly acquire an evenly distributed solution set with an acceptable time complexity. LMOPSO algorithm is proposed for multi-objective service selection problems. The proposed algorithm uses lightweight approximate distance to evaluate solutions’crowding degree and external archive to store non-dominated solutions. Simulation results prove that LMOPSO is more effective and efficient than generic algorithm NSGA-II in approximation, coverage and execution time.
Keywords/Search Tags:cloud computing, service composition, service overlaynetwork, service selection, particle swarm optimization, nichingtechnique, clustering method
PDF Full Text Request
Related items