Chuanyi Li

Energy cost minimization with job security guarantee in Internet data center

With the proliferation of various big data applications and resource demand from Internet data centers (IDCs), the energy cost has been skyrocketing, and it attracts a great deal of attention and brings many energy optimization management issues. However, the security problem for a wide range of applications, which has been overlooked, is another critical concern and even ranked as the greatest challenge in IDC. In this paper, we propose an energy cost minimization (ECM) algorithm with job security guarantee for IDC in deregulated electricity markets. Randomly arriving jobs are routed to a FIFO queue, and a heuristic algorithm is devised to select security levels for guaranteeing job risk probability constraint. Then, the energy optimization problem is formulated by taking the temporal diversity of electricity price into account. Finally, an online energy cost minimization algorithm is designed to solve the problem by Lyapunov optimization framework which offers provable energy cost optimization and delay guarantee. This algorithm can aggressively and adaptively seize the timing of low electricity price to process workloads and defer delay-tolerant workloads execution when the price is high. Based on the real-life electricity price, simulation results prove the feasibility and effectiveness of proposed algorithm. The energy cost optimization architecture is proposed for IDC operator.A heuristic algorithm is devised to select security services to guarantee the job security.The temporal diversity of electricity price is considered in minimizing the energy cost.The energy cost minimization algorithm is based on Lyapunov optimization technique.Extensive evaluation experiments demonstrate the effectiveness of our algorithms.

Process mining with token carried data

Process mining is to discover, monitor and improve real processes by extracting the knowledge from logs which are available in todays information systems. The existing process mining algorithms are based on the event logs where only the executions of tasks are recorded. In order to reduce the pre-processing efforts and strengthen the mining ability of the existing process mining algorithms, we have proposed a novel perspective to employ the data carried by tokens recorded in token log which tracks the changes of process resources for process mining in this study. The feasibility of the token logs is proved and the results of pairwise t-tests show that there is no big difference between the efforts that are taken by the same workflow system to generate the token log and the event log. Besides, a process mining algorithm (?) based on the new log is proposed in this paper. With algorithm ?, the mining efficiency as well as the mining capability is improved compared to the traditional event-log-based mining algorithms. We have also developed three plug-ins on top of the existing workflow engine, process modeling and mining platforms (YAWL, PIPE and ProM) for proving the feasibility of token log and realizing the token log generation and algorithm ?.

Software cybernetics in BPM

Interpret business process management (BPM) with the view of cybernetics.Discuss the relation between software cybernetics and BPM.Analyze the role of process discovery in software cybernetics in context of BPM.Propose a novel process discovery method based on a kind of augmented event logs.All implicit places and implicit dependencies can be found. Business Process Management (BPM) is a quickly developing management theory in recent years. The goal of BPM is to improve corporate performance by managing and optimizing the businesses process in and among enterprises. The goal is easier to achieve with the closed-loop feedback mechanism from business process execution to redesign in BPM life cycle, where the business process itself and the set of activities in BPM are viewed as a controlled object and a controller respectively. In this feedback control system, process mining plays an important role in generating feedback of process execution for redesign. However, the existing discovery methods cannot mine certain special structures from execution logs (e.g., implicit dependency, implicit place and short loops) correctly and their mining efficiencies cannot meet the requirements of online process mining. In this paper, we propose a novel discovery method to overcome these challenges based on a kind of augmented event log that will also bring new research directions for process discovery. A case study is presented for introducing how the mined model can be used in business process evolution. Results of experiments are described to show the improvements of the proposed algorithm compared with others.

Task Offloading for Scientific Workflow Application in Mobile Cloud.

Scientific applications are typically data-intensive, which feature complex DAG-structured workflows comprised of tasks with intricate inter-task dependencies. Mobile cloud computing (MCC) provides significant opportunities in enhancing computation capability and saving energy of smart mobile devices (SMDs) by offloading computation-intensive and data-intensive tasks from resource limited SMDs onto the resource-rich cloud. However, finding a proper way to assist SMDs in executing such applications remains a crucial concern. In this paper, we offer three entry points for the problem solving: first, a cost model based on the pay-as-you-go manner of IaaS Cloud is proposed; then, we investigate the problem of mapping strategy of scientific workflows to minimize the monetary cost and energy consumption of SMDs simultaneously under deadline constraints; furthermore, we consider dataset placement issue during the offloading and mapping process of the workflows. A genetic algorithm (GA) based offloading method is proposed by carefully modifying parts of GA to suit the needs for the stated problem. Numerical results corroborate that the proposed algorithm can achieve near-optimal energy and monetary cost reduction with the application completion time and dataset placement constraint satisfied.

Automatically classifying user requests in crowdsourcing requirements engineering

Abstract In order to make a software project succeed, it is necessary to determine the requirements for systems and to document them in a suitable manner. Many ways for requirements elicitation have been discussed. One way is to gather requirements with crowdsourcing methods, which has been discussed for years and is called crowdsourcing requirements engineering. User requests forums in open source communities, where users can propose their expected features of a software product, are common examples of platforms for gathering requirements from the crowd. Requirements collected from these platforms are often informal text descriptions and we name them user requests. In order to transform user requests into structured software requirements, it is better to know the class of requirements that each request belongs to so that each request can be rewritten according to corresponding requirement templates. In this paper, we propose an effective classification methodology by employing both project-specific and non-project-specific keywords and machine learning algorithms. The proposed strategy does well in achieving high classification accuracy by using keywords as features, reducing considerable manual efforts in building machine learning based classifiers, and having stable performance in finding minority classes no matter how few instances they have.

A load-aware resource allocation and task scheduling for the emerging cloudlet system

Abstract Cloudlet-assisted mobile cloud computing (MCC) emerges as a vital paradigm to address the problems of mobile services such as application time-out, data caching and traffic congestion in wireless network. The cloudlet has adequate resources to process multiple mobile requests simultaneously, but it is not as sufficient as a remote cloud data center. Currently the performance of MCC system is a subject to the lengthy network transmission latency due to the long distance between cloudlet and remote cloud. In this article, we focus on the variable user’s QoS requirements and budget of cloudlet provider, design a load-aware resource allocation and task scheduling (LA-RATS) strategy which adaptively allocates resource in MCC system for delay-tolerant and delay-sensitive mobile applications according to cloudlet’s load profile. Subsequently, a tree generation based task backfilling algorithm is proposed to raise the utilization of the cloudlet. Particularly, when cloudlet is overloaded, the restrictions of delay-sensitive applications’ deadlines are satisfied through further offloading the allocated delay-tolerant tasks in the cloudlet to distant cloud. From several systematic evaluations, it is shown that our strategy can significantly reduce the cloudlet’s monetary cost and turnaround time for delay-tolerant applications, and increase the deadline satisfaction rate of delay-sensitive applications.

Automatically Classify Chinese Judgment Documents Utilizing Machine Learning Algorithms

In law, a judgment is a decision by a court that resolves a controversy and determines the rights and liabilities of parties in a legal action or proceeding. In 2013, China Judgments Online system was launched officially for record keeping and notification, up to now, over 23 million electronic judgment documents are recorded. The huge amount of judgment documents has witnessed the improvement of judicial justice and openness. Document categorization becomes increasingly important for judgments indexing and further analysis. However, it is almost impossible to categorize them manually due to their large volume and rapid growth. In this paper, we propose a machine learning approach to automatically classify Chinese judgment documents using machine learning algorithms including Naive Bayes (NB), Decision Tree (DT), Random Forest (RF) and Support Vector Machine (SVM). A judgment document is represented as vector space model (VSM) using TF-IDF after words segmentation. To improve performance, we construct a set of judicial stop words. Besides, as TF-IDF generates a high dimensional feature vector, which leads to an extremely high time complexity, we utilize three dimensional reduction methods. Based on 6735 pieces of judgment documents, extensive experiments demonstrate the effectiveness and high classification performance of our proposed method.

Energy Cost Minimization with Risk Rate Constraint for Internet Data Center in Deregulated Electricity Markets

With the large-scale development of internet data center (IDC), the energy cost is increasing significantly and has attracted a great deal of attention. Moreover, existing scheduling optimization methods for cloud computing applications disregard the security services. In this paper, we propose a long-term energy cost minimization (ECM) algorithm with risk rate constraint for an internet data center in deregulated electricity markets. First, we formulate the stochastic optimization problem taking the temporal diversity of electricity price and risk rate constraint into account. Then, an operation algorithm is designed to solve the problem by Lyapunov optimization framework, which offers provable energy cost and delay guarantees. Extensive evaluation experiments based on the real-life electricity price demonstrate the effectiveness of proposed algorithm.

Workflow Scheduling in Grid Based on Bacterial Foraging Optimization

Optimal assignment of a workflow application in heterogeneous computing system is NP-complete in general case. We proposed the algorithm based on bacterial foraging optimization technique for Grid resource scheduling. This algorithm aims at minimizing the makespan of workflow application. To show the advantage of this algorithm, we made comparison with ant colony optimization and particle swarm optimization. The experiment shows that this bacterial foraging optimization algorithm is better than the other two algorithms in minimizing the makespan.

Construction of Microblog-Specific Chinese Sentiment Lexicon Based on Representation Learning

Sentiment analysis is a research hotspot in Nature Language Processing, and high-quality sentiment lexicon plays an important part in sentiment analysis. In this paper, we explore an approach to build a microblog-specific Chinese sentiment lexicon from massive microblog data. In feature learning, in order to enhance the quality of word embedding, we build a neural architecture to train a sentiment-aware word embedding by integrating three kinds of knowledge, including the context words and their composing characters, the polarity of sentences and the polarity of labeled words. Experiments conducted on several public datasets show that in both unsupervised and supervised microblog sentiment classification, the lexicon generated by our approach achieves the state-of-the-art performance compared to several existing Chinese sentiment lexicons and our feature learning method successfully catches both semantics and sentiment information.