Under the cloud environment, in order to take full advantage of idle time of virtual resources and meet the user's Quality of Service (QoS) requirements, a multi-dimensional QoS cloud task scheduling algorithm based on task replication was proposed. First, a cloud resource model and a user's QoS model were built. Then according to the utilization of resources and QoS satisfaction, the virtual resource with higher overall performance was chosen. Simultaneously, this algorithm duplicated a parent task in idle time to reduce the execution time. In the comparison experiments with HEFT (Heterogeneous Earliest Finish Time) and CPOP (Critical Path On a Processor), when the user's preferences perform reliability, the average reliability of the proposed algorithm was higher than that of HEFT and CPOP; when the user's preferences perform makespan and cost, the average makespan of the proposed algorithm was smaller than that of HEFT and CPOP; when the user's preferences perform nothing, the average makespan and cost of the proposed algorithm was smaller than that of HEFT and CPOP. The experimental results indicate that the proposed algorithm can improve satisfaction of customers and utilization of resources.

In order to make full use of Adjacent Reference Point (ARP) information in radio-map, a new method of establishing both location fingerprint database based on Received Signal Strength (RSS) and physical neighbor information database for each Reference Point (RP) in the off-line phase was proposed to improve the accuracy of fingerprinting-based probabilistic localization. In the on-line phase, based on the probability distribution of RSS, the system first used Bayesian inference to calculate the most adjacent points for each test point. Then, by using physical neighbor information database, the system found the physical adjacent points with respect to the most adjacent points. In the set of most adjacent and physical adjacent points, the system selected feature points for second Bayesian inference. Finally, the system estimated the position of each test point at the center of the group of feature points which had the Maximum A Posterior (MAP) probability. The simulation results show that, compared with the traditional method without physical neighbor information database, the proposed method can improve the localization accuracy by nearly 10%, which enhances the reliability of location determination.