Earticle

With the wide application of GPU in High Performance Computing, more and more heterogeneous CPU+GPU clusters have been establishedin many fields. But with the comprehensive using of heterogeneous CPU+GPU clusters, workload balancing has become an important problem when the process nodes coordinate with each other, and the execution time of a program on imbalanced clusters resides on the slowest node. Although there are many strategies and algorithms that can solve the problem of workload balancing to some extent, they generally face the problem of high consumption of communication caused by the task migration. In order to make up for the existing deficiencies, this paper proposes a virtual task migration algorithm adapted to GPU clusters on CUDA platform. This algorithm uses virtual task migration to avoid actual data transmission between nodes, so the communication overhead is obviously decreased. At last, this paper performs an actual test using matrix multiplication to verify this algorithm. The experiment results show that compared with static task partitioning, the algorithm proposed in this paper can effectively achieve dynamic workload balancing and reduce the execution time of programs on GPU clusters, thus the algorithm can significantly improve program execution performance of GPU clusters on CUDA platform.

In the near future we know that vehicles will communicate with each other to make Vehicular ad hoc network and gives the concept of intelligent transportation system. In this paper we presented the review of security in VANET. Consequently, several researchers represented the attacks and solutions in vehicular communication We analyzed some of the security problems and proposed solutions to overcome it. We discussed the need for robust Vehicular Ad hoc networks, which is very strongly dependent on their security and privacy features. This paper will review the existing attacks in VANET in the perspective approach of security.We also provided the solutions to the particular attack in VANET.

In this paper, we research on the modern intelligent healthcare platform from the perspectives of grid based cloud computing and information management system assisted Internet of things technology. In the increasing pressure of modern society, people often attend the poor general health. The more deadly is that people lack the necessary technical means that our physical condition of the related data. Hospital information construction is to adapt to the information age and the modern medical mode, necessary measures of health economy and the hospital information system is the inevitable outcome of the application of information technology in hospital. Excellent hospital information system for standardized hospital management, and security of clinical activities smoothly to carry out the relevant scientific research activity has the important meaning. Intelligent medical treatment is the use of advanced network, communication, computer and digital technology, to implement the intelligent of the medical information collection, conversion, storage, transmission and processing, and various medical business process of digital operation, thus realizes the patients and medical staff, medical institutions, medical equipment, the interaction between, gradually achieve the medical information. Intelligent medical treatment can not only improve the hospital and medical staff's working efficiency, reduce mistakes in the work, also can through the remote medical treatment, remote consultation and so on ways to solve the problem of unequal distribution of regional health resources. Our research combines the grid and cloud computing technique to build up the novel intelligent healthcare platform with IOT technique which will enhance the current medical treatment mode and level up the overall performance of the modern information management system for the general usages.

Many computer graphics rendering algorithms and techniques use ray tracing for generation of photo-realistic images, and kd-tree is the most popular acceleration data structure for ray tracing. In order to avoid the inefficient parallel performance of kd-tree construction based on surface area heuristic (SAH), an algorithm using Morton code and path compression was present in this paper. Instead of building a kd-tree layer-by-layer, the proposed approach can be performed in parallel from bottom of a conceptual perfect binary tree. Experimental results on GPU show that our work achieves a faster kd-tree construction procedure.

Cloud computing provides powerful and on-demand business environment. It is built on top of virtualized data centers. Virtualization provides the flexible infrastructure for cloud. But, managing the resources and scheduling the tasks in virtualized data center is a challenging task. This paper proposes a minimum makespan task scheduling framework named MMSF and minimum makespan task scheduling algorithm named as MMA. This algorithm is developed with two objectives. First, minimizing the total makespan and maximizing the virtual machine utilization. The task scheduling problem is formulated as multi-objective optimization problem. It is solved by using optimization techniques. Experiments shows that MMA outperforms the traditional task scheduling algorithms based on total makespan and virtual machine utilization.

Sentiment classification task has attracted considerable interest as sentiment information is crucial for many natural language processing (NLP) applications. The goal of sentiment classification is to predict the overall emotional polarity of a given text. Previous work has demonstrate the remarkable performance of Convolutional Neural Network (CNN). However, nearly all this work assumes a single word embedding for each word type, ignoring polysemy and thus inevitably casting negative impact on the downstream tasks. We extend the Skip-gram model to learn multiple sense embeddings for the word types, catering to introduce sense-based embeddings for CNN during sentiment classification. Instead of using the pipeline method to learn multiple sense embeddings of a word type, the sense discrimination and sense embedding learning for each word type are performed jointly based upon the semantics of its contextual words. We validate the effectiveness of the method on the commonly used datasets. Experiment results show that our method are able to improve the quality of sentiment classification when comparing with several competitive baselines.

Both performance and capacity of the main memory are the key to the computer systems in current architecture. DRAM, which is the most used main memory, can’t extend in capacity for its high energy consumption and repeatedly refresh. Fortunately, some new memory mediums, such as phase-change memory (PRAM), are used to replace traditional DRAM memory. These new memories have many advantages, like low energy consumption, without repeatedly refresh, high density storage, and so on. Therefore, these memories are promising. However, their low read/write performance and limited life are restricted the replacement process. In current time, hybrid memory, which consists of both PRAM and DRAM, is a good choice. In this way, the memory capacity can be extended. So, the most challenge for the hybrid memory is the performance. In this paper, we propose a dynamic hybrid LRU caching scheme (DH-LRU) for the last level cache in PRAM/DRAM hybrid main memory to improve the main memory performance. Compared with traditional cache policies, like LRU, FIFO, RANDOM, CFLRU, our DH-LRU improves performance by 4.6%. Moreover, energy consumption of write and read operation can be reduced up to 88.2%.

With the rapid development of embedded system and Internet of Things technology, embedded system and smart device based on embedded system is collecting huge amounts of data, and corresponding data processing and application method have been greatly changed, different from the traditional big data and cloud computing focus , local processing and application become important trend. This paper tries to take Cortex-A7 as the main system with the stronger ability , Hadoop distributed computing system is deployed in embedded system and could meet the demand of the future data process with directly managing the resource-constrained sensors, embedded systems and smart devices . Successfully deploying over 20GB data in the test, the system is verified that it can complete most of the functions of data processing cluster , and can also manage the collected sensors and embedded system terminal ,with better research and market promotional value.

With noise and plenty of redundant points, point cloud data obtained from non-contact scan method affects the reconstruction of geometric model. In view of the fact that de-noising and simplification of original point cloud data serve as key points of geometric reconstruction, this paper, based on noise properties, puts forward a method to eliminate noise points at different scales through combination of chord height error and improved filter operator, and introduces an algorithm to carry out non-uniform simplification of point cloud based on bounding box and curvature fusion, through which feature points can be reserved in the process of eliminating redundant point cloud, and reduced points cloud data can thus be obtained. Finally, a geometrical face model is established through integration of multi-view point cloud in different angles and data registration. It can be demonstrated through experiments that this method can improve the arithmetic speed with the reconstruction integrity guaranteed and establish highly realistic 3D face model with structural integrity and smooth surface.

K-Means is a widely used partition based clustering algorithm famous for its simplicity and speed. It organizes input dataset into predefined number of clusters. K-Means has a major limitation -- the number of clusters, K, need to be pre-specified as an input. Pre-specifying K in the K-Means algorithm sometimes becomes difficult in absence of thorough domain knowledge, or for a new and unknown dataset. This limitation of advance specification of cluster number can lead to “forced” clustering of data and proper classification does not emerge. In this paper, a new algorithm based on the K-Means is developed. This algorithm has advance features of intelligent data analysis and automatic generation of appropriate number of clusters. The clusters generated by the new algorithm are compared against results obtained with the original K-Means and various other famous clustering algorithms. This comparative analysis is done using sets of real data.

Virtual machine resource allocation problem is closely related to the resource requirement of physical host, and a novel configuration method of virtual machine is proposed. First, the system analyzes the resource requirements of the related virtual machine to the physical host. Secondly, with the help of the similarity between the virtual machine, the fitness is configurated. Secondly, based on the degree of adaptation, the scheduling of the related virtual machine is completed. Two typical methods are selected and compared with the method in this paper. The experimental results confirm the effectiveness of the new method.

In cloud computing era, automatic parallelization is still significant for virtualization platform. However, after several decades of development, the overall effect is still to be improved. Summary of the mainstream technology developments will be beneficial to reveal the future direction and trend. This paper reviews the technology of loop parallelization, which is the key issue in automatic parallelization. After introducing the basic models and approaches, we focus on the recent developments, on which we obtain the trend of this field and the conclusions about future.

With the rapid development of national economy, the demand for electricity is more intense. There is a new demand for the security, reliability and applicability of the power distribution system. The power distribution system is an important component of power systems. It plays an important role in connecting power production and users. Now the most important problem is nonlinear combinatorial optimization in large-scale mixed lines and how to reconfigure the distribution network. However, by deeply analyzing and studying immune particle swarm optimization, the mathematical model of distribution network reconfiguration is established. By coding the characteristics of closed-loop design and open-loop operation, it aims to reduce the consumption of network, balance the grid load, eliminate the current overload and achieve the ultimate goal of improving the power supply reliability.

When considering the coexistence of Software-Defined Networking (SDN) and non-SDN, unified management of SDN calls for study on both the information representation and the communication pattern. This paper first proposes the use of the extension theory in order to formally define management information of SDN in a unified manner. And this paper then introduces the concept of unified manager to the communication pattern for unified management of SDN, and considers application of the queuing theory from the perspective of the basic queue and the queue spaces. Finally, simulation results validate that, proposed approach based on the queuing theory is feasible for unified management of SDN from the viewpoint of the communication pattern.

The conventional operating system usually works on a particular workstation but the same when implemented on a distributed environment like cloud faces many problems. This paper analyzes and explains how does the cloud operating system works in order to resolve these problems. It also differentiates the working mechanism of both based on basic storage management service provided by the cloud. At the end it provides the issues and the future challenges faced in the cloud operating system.

Finding k nearest neighbours to one query point on road network is the typical kNN problem, and attracts lots of research interests in recent years with the development of mobile technologies. In this paper, we study a variant of the kNN problem, namely mkNN, which returns k objects for m query points. We firstly propose a baseline algorithm, which solves the mkNN problem in naive way. We then propose an optimized algorithm, which avoids the redundant computation of the baseline algorithm by organizing objects using index to firstly get candidate objects, such that to reduce the number of visited nodes of the given road network. Our third algorithm further improves the performance by using a tighter threshold to greatly reduce the number of visited nodes. We make detailed comparison between the three algorithms and show their benefits w.r.t. the mkNN problem.

For the security issue of the single tag ownership transfer, this paper proposes a multi-owner RFID Tag ownership transfer protocol based on the Chinese remainder theorem. By using secret sharing scheme, this protocol divides the sharing key into n parts, which then are distributed to n owners. Only when t owners cooperate together, the sharing key will be recovered. Thus, it ensures the tag’s security during the authentication and transformation of the ownership. Under the Linux environment, this protocol was achieved through simulationtest and the experiment data about the calculate time of the tag was got. Then with the number of owners increasing gradually, different results were compared. The result shows that in the protocol, when increasing the owners’ number, the calculate time of tag is in the acceptable range, which is also apply to the low cost tag.

This paper mainly using the supplies distribution model in emergency situation, conducted a in-depth research on the problem of supplies distribution in the disaster relief, including single supplies and multiple supplies respectively with and without transport capacity constraints, and then implement its calculation through computer program; next, emergency supplies distribution model of multiple target and dynamic conditions is discussed in further, that is to say, considering some conditions such as the road condition change in the existing logistics system when disasters occur, in this case, the disaster relief supplies distribution model of object-oriented is introduced to guarantee the supplies. At present, there is still no systematic disaster relief logistics system in our country, so the study results of this paper have a certain reference value to the build and implementation of disaster relief logistic system in our country.

Network virtualization is an important technology that allows multiple heterogeneous networks running simultaneously share substrate network, it can effectively solve the problem of network ossification. Earlier research has made many heuristic algorithms, but it does not consider the time factor in the embedding process, existing algorithm often make decision according to the present available substrate network resource. This paper, we first explore the time and space of the two dimensional network model, form the perspective of the two-dimensional space-time, we proposed two-dimensional discrete weighted method and multi commodity flow virtual network embedding with time-varying model. We also proposed a virtual network algorithm based on space–time. The simulation result show that the proposed algorithm can effectively utilize the network resources and make substrate network resources maintain the load balance for a long term.

In the study of reasons that cause the flush illegal operating vehicles, we found out that the insufficient legal operating vehicles were the main cause. In order to efficiently solve the illegal operating, the paper brings out the research of scheduling algorithm about the legal operating vehicles based on NFC. The research uses the NFC technology to realize the intelligent and fast identification and the scheduling algorithm to realize the problem of insufficient legal operating vehicles. The scheduling can be divided into two stages--the multi-level queues high response ratio first scheduling and the multi-level queues earliest available time first scheduling. The algorithm can choose the user whose response ratio is the highest from the users’scheduling list and allocate the user who has the highest ratio to the legal operating vehicle who has the earliest available time that in the same area. Experiment indicates that the scheduling success ratio, throughput, average waiting time of the scheduling algorithm are superior to random travel of the drivers and can efficiently resolve the insufficient of legal operating vehicles.

As the use of wireless applications is increasing every day; it results into shortage of limited available spectrum. Licensed spectrum is unutilized and unlicensed spectrum is overcrowded, so to increase the efficiency of allocated spectrum and to fulfill the need of unlicensed spectrum, CRN is proposed. Some routing protocols are designed for wireless communication keeping in mind that all nodes are cooperating. So this approach can solve the problem of spectrum scarcity. However due to the flexibility of such network, it is vulnerable to many security threats. Jamming attack is one of the most popular security threats which cause denial of service in the network. Various solutions are available in the literature to deal with jamming attack but still there are some open challenges in jamming attack. In this paper various anti-jamming techniques and analyze the open issues of jamming which are still there in the cognitive radio network. An anti-jamming technique is proposed in this paper to deal with the open issues. Simulation is done in MATLAB-2013 and results show that proposed method performs well.

Recommender systems provide users with personalized suggestions for products. A key challenge is how to improve the diversity of recommendation results as much as possible, while maintaining reliably accurate suggestions. Although the bipartite graph based probabilistic spreading algorithm has its advantages of good accuracy and low computational complexity, its diversity is poor. In this paper, we introduce a cost-aware probabilistic spreading algorithm, and show how it can improve both recommendation accuracy and diversity by designing different spreading costs. Comparative experiments on widely used datasets confirm the effectiveness of the cost-aware probabilistic spreading approach in terms of accuracy, aggregate diversity and individual diversity of recommendation results. In addition, the time complexity of the proposed algorithm is also analyzed.

Traditional wind power prediction is only applicable to a single wind farm. Aim at this isolated prediction method. In this paper, combing with the information sharing and Interconnection mechanism of energy Internet, we propose an output power prediction method for multiple wind farms based on DBPSO-LSSVM model. Firstly, collect SCADA data of multiple wind farms in different areas. Secondly, delete outliers of different farms based on DBSCAN algorithm and select multiple wind fields training samples. And searching the optimal input parameters of LSSVM based on particle swarm algorithm to construct every wind farm model. Thirdly, predict multiple wind fields power combined with numerical weather prediction system. The method we propose can be used to make the scheduling plan in advance to solve a large number of abandoned wind power rationing problem every year. In experiment, the method we propose has the lowest error rate compares to LSSVM and BP-neural network. It’s more suitable to predict wind fields in different areas.

In current years, data streams have been gradually turn into most important research area in the field of computer science. Data streams are defined as fast, limitless, unbounded, river flow, continuous, stop less, massive, tremendous unremitting, immediate, stream flow, arrival of ordered and unordered data. Data streams are divided into two types, they are online and offline streams. Online data streams are mainly used for real world applications like face book, twitter, network traffic monitoring, intrusion detection and credit card processes. Offline data streams are mainly used for manipulating the information which is based on web log streams. In data streams, data size is extremely huge and potentially infinite and it is not possible to lay up all the data, so it leads to a mining challenge where shortage of limitations has occur in hardware and software. Data mining techniques such as clustering, load shedding, classification and frequent pattern mining are to be applied in data streams to get useful knowledge. But, the existing algorithms are not suitable for performing the data mining process in data streams; hence there is a need for new techniques and algorithms. The main objective of this research work is to perform the clustering process in data streams and detecting the outliers in data streams. New hybrid approach is proposed which combines the hierarchical clustering algorithm and partitioning clustering algorithm. In hierarchical clustering, CURE algorithm is used and enhanced (E-CURE) and in partitioning clustering, CLARANS algorithm is used and enhanced (E-CLARANS). In this research work, the two algorithms E-CURE and E-CLARANS are combined (Hybrid) for performing a clustering process and finding the outliers in data streams. The performance of this hybrid clustering algorithm is compared with the existing hybrid clustering algorithms namely BIRCH with CLARANS and CURE with CLARANS. The performance factors used in this analysis are clustering accuracy and outlier detection accuracy. By analyzing the experimental results, it is observed that the proposed hybrid clustering approach E-CURE with E-CLARANS performance is more accurate than the existing hybrid clustering algorithms.

Cloud Resource Providers (CRPs) can form CRPs coalitions for executing large scale application tasks by pooling their resources together in cloud computing. However, forming such CRPs coalitions is a challenging problem especially when the trust relationships among CRPs have to be taken into account. Aiming at this problem, we model the formation of CRPs coalitions by considering the trust and reputation of the participanting CRPs in clouds in this paper. An algorithm TCFA for coalition formation is proposed, which can enable the CRPs with high reputation to join into a coalition reducing the cost of execution and guaranteeing the maximum profit for the participanting CRPs. Furthermore, we prove that the formed coalition determined by TCFA is stable, which means that the CRPs that are the part of the coalition do not have incentives to break away from it. Meanwhile, we prove that TCFA can produce a Pareto optimal solution. The results of simulation experiments show that our algorithm TCFA not only can produce stable coalitions structure composed of CRPs with higher reputation, but also can obtain higher individual profit among coalitional members.

With the development of cloud computing technology, people not only want to pursue the shortest time to complete the tasks by using cloud computing, but also hope to take into the running costs of machines. Existing task scheduling algorithm in the cloud computing environment has been unable to meet people's needs. As an extension and generalization of the model checking theory, probability model checking is also used in many fields, such as random distributed algorithm and other areas. The task scheduling algorithm based on the particle swarm optimization algorithm combined with probability model is proposed in this paper. The algorithm defines the fitness functions of the time cost and the running cost. The fitness functions can improve the efficiency of the cloud computing platform. At the same time, the probability model can be used to analyze the running states of machines and the computing capability of the nodes in the cloud cluster. The probability, which is calculated by the probability model, provides the basis for changing particle swarm algorithm’s the inertia factor and the learning factor, so as to solve the drawback that the inertia factor and the learning factor solely depend on the fixed value.

Modern cloud data centers are virtualized for resource multiplexing and services consolidations. Virtual machines (VMs) residing in the same server cluster share the same hardware resources and power supply while they may have different QoS requirements for their services and applications. Moreover, the power consumption of the server cluster is highly dynamic since different virtual machines have different workloads due to different services requests. Cluster level power and QoS coordination is crucial for data center level energy efficiency coordination as well as high quality service provisioning. In this paper we propose the power and QoS aware multi-level resource coordination and scheduling in virtualized servers, i.e., the cluster-level power control layer, the VMs resource allocation layer, and the QoS optimization layer. This three-level controlling framework schedules and allocates hardware resource for QoS guarantee and cluster level power management. We use dynamic frequency scaling for QoS mitigating when power budget changes. The experiment results show that the proposed multi-level coordinated control architecture consumes 5.36% and 6.96% less power for web servers and computing intensive virtual machines, respectively while it can guarantee the response time of web server and execution time of computing tasks no more than those without the proposed controlling approach.