Abstract:Energy efficiency of cloud data centers has received significant attention recently as data centers often consume significant resources in operation. Most of the existing energy-saving algorithms focus on resource consolidation for energy efficiency. Accurate energy consumption model is the basis for these algorithms. This paper proposes an accurate energy model to predict energy consumption of single machine. In most of the existing cloud computing energy studies, linear models are used to describe the relationship between energy consumption and resource utilizations. However, with the changes in computer architecture, the relationship between energy and resource utilizations may not be linear. In fact, this paper explored a variety of regression analysis methods to estimate the energy consumption accurately while using low computational overhead. Initially, multiple linear regression models are used, but they often do not produce good enough results. Afterwards, this paper chooses three non-linear models and finally settled with the polynomial regression with Lasso as it produces the best estimation. Experimental results show that in adoption of energy model presented in this paper, the prediction accuracy can reach more than 95%.

Abstract:In recent years, the huge resource consumption problem of data centers is being widely concerned. Virtual machine monitor (VMM) can consolidate virtual machines (VMs) onto fewer servers via VM migration to improve the energy efficiency of data centers. This paper surveys the recent works on energy-efficient VM consolidation, and summarizes three research challenges. Among them, this work considers the resource consumption overhead caused by the waiting of virtual machines for server resource scheduling. The study theoretically and experimentally proves that under realistic constraints, this overhead remains steady as the number of consolidating VMs grows. Experiments based on a representative benchmark show that, on average, 11.7% of the server's CPU resource is occupied by the overhead. In addition, in order to fill in the gap on existing approaches, this paper proposes margin reserved consolidation (MRC) algorithm. Simulation results show that MRC outperforms the state of the art baseline in terms of server resource violation probability.

Abstract:Main memory accounts for a large and increasing fraction of the energy consumption in multi-core systems. Therefore, it is critical to address the power issue in the memory subsystem. This paper presents a solution to improve memory power efficiency through coordinating page allocation and thread group scheduling (CAS). Under the proposed page allocation, all threads are partitioned into different thread groups, where threads in the same thread group occupy the same memory rank. Thread group scheduling is then implemented by adjusting default Linux CFS. The CAS alternates active partial memory periodically to allow others power down and prolongs the idle ranks. Experimental results show that this approach improves energy saving by 10% and reduces performance overhead by 8% comparing with the state of the art polices.

Abstract:Network virtualization will be an enabler for intelligent energy-aware network deployment. Current networks are designed for peak loads, resulting in inadequate resource utilization and energy consumption waste. Due to current power consumption insensitiveness of network equipment to traffic load, resource consolidation becomes an effective energy-saving technology. Based on the virtual network mapping characteristics and the substrate network energy consumption, this paper presents a multi-objective decision-making model that is also a mixed integer programming model for energy efficient virtual network embedding. To address high time complexity in solving the mixed integer programming model, the paper analyzes the dynamic characteristics of the virtual network mapping, constructs virtual network mapping dictionary database and proposes a method for training substrate network resource utilization, as well as an algorithm which actively hibernates the substrate nodes and links. By this method, the virtual network can be embedded in a smaller set of substrate nodes and links, which helps to increase the number of hibernating substrate nodes and links, and achieve energy-effective virtual network mapping. Simulation results demonstrate the proposed method can effectively improve the number of hibernating nodes and links of substrate network, and significantly reduce energy consumption of substrate network.

Abstract:With the development of the Internet, the scale of data center increases dramatically. How to analyze the data stored in the data center becomes the hot research topic. Programmers resort to the machine learning to analyze unstructured or semi-structured data. Thus, energy efficient machine learning is crucial for green data centers. Based the observation that there is redundant computation in the machine learning applications, this paper proposes a system which can save the power usage by removing the redundant computations and reusing the previous computation results. Evalution shows that for the typical k-means and PageRank applications the presented algorithm results 23% and 17% power saving.

Abstract:In order to reduce both electricity bills and carbon emission, data center operators begin to build their own on-site green energy plants. However, challenges arise with the fluctuating workload, temporally diverse electricity price and intermittent green energy. To deal with these challenges, this paper presents an online workload scheduling algorithm which can minimize the total electricity bills of a data center without any future information about workload, electricity price or green energy availability. First, a model for the total electricity bills of a data center is introduced. Then a stochastic optimization problem to minimize the electricity bills is formulated. Finally, solution to the optimation problem is made to form the corresponding workload scheduling policy. Experimental results based on real-world traces show that the proposed algorithm can effectively reduce the total electricity bills while guaranteeing the workload performance.

Abstract:With the recent development on heterogeneous hardware, heterogeneous parallel programming model has been widely used with the intension of simplifying programming and improving efficiency. This paper analyses latest achievements in heterogeneous parallel programming interfaces and runtime supporting systems, and solutions to new problems brought by heterogeneous architectures and various applications. In the end, some future trends in this area are discussed.

Abstract:Due to the diversity of resources and different management requirements, cloud management is faced with great challenges in complexity and difficulty. For constructing a management system to satisfy a specific management requirement, redeveloping a solution based on existing management system is usually more practicable than developing the system from scratch. However, the difficulty and workload of redevelopment are also very high. In this paper, a runtime model based approach is presented to managing diverse cloud resources. First, the runtime model is constructed for each type of cloud resources based on their management interfaces. Second, the composite runtime model is build for all managed resources through merging their runtime models. Third, cloud management is setup to meet specific requirements through model transformation from the composite model to the customized models. Additionally, based on OpenStack and Hyperic, a runtime model based management system is implemented to manage the hardware and software resources of virtual machines with the proposed approach. The results prove that new approach is feasible and effective.

Abstract:The current test case reduction methods can not improve the effectiveness of fault localization, and the current fault localization approaches do not fully analyze the dependency of program elements. To solve these problems, this study proposes an automatic fault localization approach combining test case reduction and joint dependency probabilistic model. Different from the usual test case reduction approach, the failed test cases are fully considered in the proposed test cases reduction method based on execution path in order to provide effective test cases for fast and accurate fault localization. This paper defines a novel statistical model—Joint dependency probabilistic model. In this model, the control dependency and data dependency between program elements, the execution states of each statement are analyzed. An automatic fault localization approach is presented based on joint dependency probabilistic model. It ranks the suspicious statements by calculating the joint dependency suspicion level of the statement. Experimental results show that this approach is more effective than current state-of-art fault-localization methods such as SBI, SOBER, Tarantula, and RankCP.

Abstract:With its rising popularity, as evidenced in social networks, online shopping platforms and email systems, detection of Web spammer has already become one of the hottest topics in the data mining field. The main challenge of Web spammer detection is how to recognize spammer behavior patterns by examining spammer features and attributes from big dataset in order to limit the proliferation of Internet spam and insure quality of Internet service. This paper presents an overview of Web spammer detection, along with a comparison over the difference between traditional and burgeoning spammer detection approaches. The key techniques and evaluation methods are classified and discussed from several aspects. At last, the prospects for future development and suggestions for possible extensions are emphasized.

Abstract:IP geolocation aims at determining the geographic location of an Internet host, which can improve the performance and security of the Internet application, and bring about novel services. This paper firstly illustrates the concept and applications of the IP geolocation, and then categorizes current typical IP address geolocation algorithms into two classes, namly, client-independent geolocation algorithms and client-dependent geolocation algortithms. Next, the main ideas of the representative algorithms of each class are illustrated, and the privacy protection techniques and the influence of new techniques are discussed. Finally, a comprehensive comparison is made on the IP geolocation algorithms and systems, and the future trends of the IP geolocation are discussed.

Abstract:Opportunistic routing (OR) significantly improves transmission reliability and network throughput in wireless mesh networks by taking advantage of the broadcast nature of the wireless medium. With network coding (NC), OR can be implemented in a simple and practical way without resorting to a complicated scheduling. With the introduction of NC, how to reduce redundant transmission of coded packets becomes a very important problem in OR protocol. MORE, et al. protocols estimate the expected number of transmissions for each forwarder based on periodic measurements of the average link loss rates. However, these approaches may suffer severe performance degradation in dynamic wireless environments. Recently, some studies, known as CCACK, employ orthogonal vector as feedback to reduce redundant transmission of coded packets. The analysis of CCACK scheme indicates that the false-positive probability is reduced at the cost of increasing the false-negative probability, which results in unnecessary packets transmission. This paper proposes a NC-based OR protocol, named CFACK, based on cumulative coding coefficient feedback acknowledgement. In this scheme, the coding vectors piggybacked in coded packets are used as feedback information, and each forwarder overhears coding vectors sent by downstream nodes. Through correlation analysis between coding vectors from upstream nodes and downstream ones each forwarder knows whether its knowledge space is covered by its downstream nodes. This paper proves that CFACK is completely free from any false-positive and false-negative problem in reliable network. The efficiency of CFACK in unreliable network is also analyzed, and the result shows that in random topologies embedding an extra ACK vector in each packet can guarantee 90% accuracy. Evaluation shows that, compared with CCACK, CFACK significantly improves throughput by reducing unnecessary packet transmission, with average improvements of 72.2%. Furthermore, the overheads of encoding computation, storage, and header of CFACK are less than that of CCACK.

Abstract:As the most important step in shape-based image retrieval, the description of image contour should reflect the information of global shape and key points, and be robust to random noise. This paper proposes a new image retrieval method based on contour reconstruction and feature point chord length. First, the contour of the shape is extracted, and in order to reduce the distortion caused by random noise, the contour is reconstructed by analyzing the energy retention rate. Then, base on the new defined supportive region, the feature intensity is calculated at each point of the contour to extract the valid feature points. After that, the contour feature function is structured by using the chord length between contour points and corresponding feature points. Finally, the shape descriptors are processed to meet the invariance property. A significant amount of experiments show that, in both normal and noisy sample sets, the proposed method demonstrates better performance compared with other seven techniques.

Abstract:An image clustering segmentation model combined with variational level set and fuzzy clustering is proposed in this paper. An external fuzzy clustering energy based on the local image information and a new regularization energy with respect to the zero level set are introduced in the energy functional, which makes the proposed model robust in noisy image segmentation. An internal energy that forces the level set function to be close to a signed distance function is introduced in the energy functional, which can completely eliminate the need of the expensive periodical re-initialization procedure for level set function during its evolution. Furthermore, this paper proposes a variational formulation to update the cluster centers in the procedure of clustering, which realizes the semi-supervised clustering segmentation. The experimental results show that the proposed model, compared with the FCM clustering model, CV model and Samson model, can reduce the influence of noise and get better segmentation results for different kinds of images.

Abstract:To overcome the shortcomings of descriptor BRIEF which is sensitive to image rotation, this paper proposes a improved descriptor RIBRIEF which has the advantages of good identification ability, high descriptor extraction speed, less memory usage, strong robustness and rotation invariant. The study shows that real-time performance of image matching algorithm is largely decided by the number of feature points, the search times of matching points and the computational complexity of descriptor similarity. It therefore proposes optimization algorithms to improve real-time performance of image matching by combining descriptor index and descriptor cluster, applying FAST to stable feature point extraction and calculating descriptor similarity with logic operations. Compared with SURF and BRIEF, experimental results show that RIBRIEF has better performance in robustness and real-time.

Abstract:Tracking of abrupt motion is a challenging task in computer vision due to the large motion uncertainty induced by camera switching, sudden dynamic change, and rapid motion. This paper proposes an ordered over-relaxation Hamiltonian Markov chain Monte Carlo (MCMC) based tracking scheme for abrupt motion tracking within Bayesian filtering framework. In this tracking scheme, the object states are augmented by introducing a momentum item and the Hamiltonian dynamics (HD) is integrated into the traditional MCMC based tracking method. At the proposal step, the ordered over-relaxation method is adopted to draw the momentum item in order to suppress the random walk behavior induced by Gibbs sampling. In addition, the paper provides an adaptive step-size scheme to simulate the Hamiltonian dynamics in order to reduce the simulation error. The proposed tracking algorithm can avoid being trapped in local maxima with no additional computational burden, which is suffered by conventional MCMC based tracking algorithms. Experimental results reveal that the presented approach is efficient and effective in dealing with various types of abrupt motions compared with several alternatives.

Abstract:Steganography payload is one of the four key performance indicators for information hiding. Previous research on information hiding has mainly focused on the other three performance indicators, namely, the robustness, transparency and computational complexity, with very few work being carried out regarding to the steganography payload. This research aims to effectively improve the theoretical system of information hiding. According to the JPEG2000 compression standard and the human eye sensitivity of changing wavelet coefficients, along with the aid of distortion cost function, the study discriminates the wavelet coefficients' carrying capacity of secret information: The smaller the distortion cost function value, the stronger the wavelet coefficients' carrying capacity. Conversely, the larger distortion cost function value, the weaker the wavelet coefficients' carrying capacity. When the distortion cost function value is greater than one, the coefficient will not have enough capacity to carry information, i.e., wet coefficients. By means of the maximum steganography payload experiments along with the bit full embedding, over bits embedding and wet embedding experiments, the effectiveness of the proposed estimation method in this work is verified.