Workflow has proven to be a highly effective computing model for a variety of scientific applications, offering flexible data types and unstructured parallelism that surpasses simple parallel execution models such as MapReduce. However, current workflow management systems in cloud computing environments experience unnecessary delays in task execution due to the separation of task execution and data transfer processes, which causes a child task to wait until all its predecessor tasks complete, rather than waiting only for necessary input data becoming ready. The goal of this paper is to eliminate the unnecessary delay of child tasks in a workflow, which is achieved through a new workflow engine architecture that separates workflow planner from workflow executor in the general framework of the DATAVIEW scientific workflow management system. This new engine architecture can be generalized and applied to other workflow systems. Our design integrates a new task release mechanism based on a data dependency model with the workflow executor of DATAVIEW. This approach enables prompt task launching once input data becomes available, instead of waiting for all predecessor tasks to finish. The architecture employs distributed algorithms for implementing the workflow executor and the task executors, performing various optimization on data movement, task movement, and communication among different subsystems. The experiments show that our new architecture based on the new task release model can significantly reduce overall execution time of a workflow in DATAVIEW.

Scientific workflow has become a popular cyberinfrastructure paradigm to accelerate scientific discoveries by enabling scientists to formalize and structure complex scientific processes. With the recent success of deep learning models in many scientific applications, there is a rising need for infrastructure-level support for deep learning technologies in scientific workflow cyberinfrastructures. However, current scientific workflow cyberinfrastructures and GPU-enabled deep learning frameworks are developed separately, neither alone can be a satisfactory choice. In this paper, We propose the Deep-Learning-as-a-Workflow approach in DATAVIEW, which for the first time incorporates native infrastructure level support for GPU-enabled deep learning in a scientific workflow management system and enables the fast training and execution of neural networks as workflows (NNWorkflows) leveraging various types of GPU resource configurations. Our experiments demonstrate the salient usability feature of DATAVIEW in providing seamless infrastructure-level support to both scientific and deep learning workflows in one system, while delivering competitive (better in most cases) learning efficiency compared to the conventional implementations based on Keras.

Although Virtual Machine placement has been extensively researched in traditional Cloud Computing environments, it remains an open challenging problem for Volunteer Cloud Computing, which exhibits several divergent characteristics, including intermittent availability of nodes and unreliable infrastructure. In this paper, we model the Virtual Machine placement problem in Volunteer Cloud Computing as a bounded 0-1 multi-dimensional knapsack problem and develop three heuristic based algorithms to meet the objectives and constraints specific to Volunteer Cloud Computing. Empirical evidences on a real Volunteer Cloud Computing test-bed show the competitive performance results of these algorithms.

The Massive Open Online Courses (MOOCs) have emerged as a great resource for learners. Numerous challenges remain to be addressed in order to make MOOCs more useful and convenient for learners. One such challenge is how to automatically extract a set of keyphrases from MOOC video lectures that can help students quickly identify a suitable knowledge without spending too much time and expedite their learning process. In this paper, we propose SemKeyphrase, an unsupervised cluster-based approach for keyphrase extraction from MOOC video lectures. SemKeyphrase incorporates a new ranking algorithm, called PhaseRank, that involves two phases on ranking candidate keyphrases. Experiment results on a real-world dataset of MOOC video lectures show that our proposed approach outperforms the state-of-the-art methods by 16% or more in terms of F1 score.

The task and worker recommendation problems in crowdsourcing systems have brought up unique characteristics that are not present in traditional recommendation scenarios, i.e., the huge flow of tasks with short lifespans, the importance of workers’ capabilities, and the quality of the completed tasks. These unique features make traditional recommendation approaches no longer satisfactory for task and worker recommendation in crowdsourcing systems. In this article, we propose a two-tier data representation scheme (defining a worker--category suitability score and a worker--task attractiveness score ) to support personalized task and worker recommendations. We also extend two optimization methods, namely least mean square error and Bayesian personalized rank, to better fit the characteristics of task/worker recommendation in crowdsourcing systems. We then integrate the proposed representation scheme and the extended optimization methods along with the two adapted popular learning models, i.e., matrix factorization and kNN, and result in two lines of top- N recommendation algorithms for crowdsourcing systems: (1) Top- N -Tasks recommendation algorithms for discovering the top- N most suitable tasks for a given worker and (2) Top- N -Workers recommendation algorithms for identifying the top- N best workers for a task requester. An extensive experimental study is conducted that validates the effectiveness and efficiency of a broad spectrum of algorithms, accompanied by our analysis and the insights gained.

Alternative cloud systems that depend on spare resources of volunteer computers are emerging, complementing the conventional data center based clouds. These alternative cloud systems have one characteristic in common - they do not rely on dedicated data centers to provide the cloud services. Delivering reliable cloud services over volatile donated computing resources of volunteer hosts is one of the main challenges in such opportunistic cloud systems that run over scavenged resources. This paper discusses the design of a multi-state semi-Markov process based availability and reliability prediction model for nodes in volunteer cloud computing systems. It also presents the implementation and integration of the model in a real volunteer cloud system, cuCloud. Moreover, the evaluation and empirical results obtained from the experimentation performed on real cuCloud system is discussed.

Current Cloud Computing is primarily based on proprietary data centers, where hundreds of thousands of dedicated servers are setup to host the cloud services. In addition to the huge number of dedicated servers deployed in data centers, there are billions of underutilized Personal Computers (PCs), usually used only for a few hours per day, owned by individuals and organizations worldwide. The vast untapped compute and storage capacities of the underutilized PCs can be consolidated as alternative cloud fabrics to provision broad cloud services, primarily infrastructure as a service. This approach, thus referred to as "no data center" approach, complements the data center based cloud provision model. In this paper, we present our opportunistic Cloud Computing system, called cuCloud, that runs on scavenged resources of underutilized PCs within an organization/community. Our system demonstrates that the "no data center" solution indeed works. Besides proving our concept, model, and philosophy, our experimental results are highly encouraging.

With the increased popularity of cloud computing, the number and scales of cloud data centers have kept growing at unprecedented speeds. In the meanwhile, the energy consumption by the data centers has kept commensurately increasing as well. Therefore, the focus of cloud resource management and scheduling has relatively shifted from mere performance to also energy efficiency. In this paper, we present a novel, Energy-Aware Task Scheduling framework that makes integrated exploitation of the two well-known energy saving techniques, DVFS and VM Reuse, on cloud task scheduling in a data center. We present our scheduling approach and framework via a specific algorithm, called EATS-FFD, that assumes FFD as its base scheduling policy. With minor modification, the presented framework can be made to work with a different base scheduling policy, resulting in a correspondingly different scheduling algorithm. Our approach achieves better energy-efficiency without sacrificing system QoS. The effectiveness of our approach is evaluated under various experimental scenarios using the Cloud Report tool running on the open source CloudSim platform.

As one of the important problems in molecular biology, motif search is computationally expensive, especially when the size of DNA sequences is large. Extended from a graduate course project in parallel and distributed computing (PDC), this paper investigates two different programming frameworks, namely MapReduce and MPI on motif finding. We implemented a serial algorithm, a MapReduce based algorithm, and a MPI program to calculate the best motif in given DNA sequences. The experimental results demonstrate that our MPI program outperformed both the MapReduce-based algorithm and the serial program with superior efficiency.

We are motivated to study (and eventually deliver) a cloud facility that has its "back door" open for incorporating voluntary computing resources, and such implemented cloud infrastructures have the potential to harmoniously serve both on-demand and non on-demand requests/jobs at possibly the lowest cost because of involvement of the low or no cost of using voluntary computing resources. This short paper presents our idea of enhancing the "backfill computing" model originally proposed by Marshall, et al. by incorporating voluntary computing resources into the resource pools of clouds, and summarizes the simulation results we had recently obtained.

Fast and accurate estimations for complex queries are profoundly beneficial for large databases with heavy workloads. In this research, we propose a statistical summary for a database, called CS2 (Correlated Sample Synopsis), to provide rapid and accurate result size estimations for all queries with joins and arbitrary selections. Unlike the state-of-the-art techniques, CS2 does not completely rely on simple random samples, but mainly consists of correlated sample tuples that retain join relationships with less storage. We introduce a statistical technique, called reverse sample, and design a powerful estimator, called reverse estimator, to fully utilize correlated sample tuples for query estimation. We prove both theoretically and empirically that the reverse estimator is unbiased and accurate using CS2. Extensive experiments on multiple datasets show that CS2 is fast to construct and derives more accurate estimations than existing methods with the same space budget.

Twig pattern matching is a critical operation for XML query processing, and the holistic computing approach has shown superior performance over other methods. Since Bruno et al. introduced the first holistic twig join algorithm, TwigStack, numerous so-called holistic twig join algorithms have been proposed. Yet practical XML queries often require support for more general twig patterns, such as the ones that allow arbitrary occurrences of an arbitrary number of logical connectives (AND, OR, and NOT); such types of twigs are referred to as B-twigs (i.e., Boolean-Twigs) or AND/OR/NOT-twigs. We have seen interesting work on generalizing the holistic twig join approach to AND/OR-twigs and AND/NOT-twigs, but have not seen any further effort addressing the problem of AND/OR/NOT-Twigs at the full scale, which therefore forms the main theme of this paper. In this paper, we investigate novel mechanisms for efficient B-twig pattern matching. In particular, we introduce “B-twig normalization” as an important first-step in our approach toward eventually conquering the complexity of B-twigs, and then present BTwigMerge-the first holistic twig join algorithm designed for B-twigs. Both analytical and experimental results show that BTwigMerge is optimal for B-twig patterns with AD (Ancestor-Descendant) edges and/or PC (Parent-Child) edges.

While an XML database consists of a collection of data trees, an XML Query is essentially a tree pattern associated with selection predicates. Various structural join algorithms have been designed to obtain the matches of a tree pattern within an XML database, and more recently, holistic twig joins were proposed as better alternatives to structural joins. The proposed twig join algorithms typically rely on a two-phase processing scheme: the first phase discovers the root-to-leaf paths (solutions) and the second phase join-merges the identified paths for producing the desired matches of a twig pattern. In this paper, we address the deficiencies (especially, the path merging deficiency) as found in TwigStack - the representative of twig join algorithms; we then present an alternative algorithm, called MyTwigStack, that improves TwigStack through an effective path merging scheme.

As the popularity of XML keeps growing rapidly, XML compliant structured document management becomes an interesting and compelling research area. Query optimization for structured documents stands out as a very challenging issue because of the much enlarged optimization (search) space, which is a consequence of the intrinsic complexity of the underlying data model of structured documents. We therefore propose to apply deterministic transformations on query expressions to most aggressively control the search space and fast achieve a sufficiently improved alternative (if not the optimal) for each incoming query expression. This idea is not just exciting but practically attainable. This paper first provides an overview of our optimization strategy, and then focuses on the implementation issues of our transformation system for structured document query optimization.

In this paper, a new technique is developed to support the query relaxation in biological databases. Query relaxation is required due to the fact that queries tend not to be expressed exactly by the users, especially in scientific databases such as biological databases, in which complex domain knowledge is heavily involved. To treat this problem, we propose the concept of the so-called fuzzy equivalence classes to capture important kinds of domain knowledge that is used to relax queries. This concept is further integrated with the canonical techniques for pattern searching such as the position tree and automaton theory. As a result, fuzzy queries produced through relaxation can be efficiently evaluated. This method has been successfully utilized in a practical biological database - the GPCRDB.

The number of documents published via the World Wide Web in the form of SGML/HTML has been rapidly growing for years. Efficient, declarative access mechanisms for this type of document-structured documents in general-are becoming of great importance. This paper reports our most recent advance in pursuit of the effective processing and optimization of structured document queries, which are important for large repositories of structured documents. Our methodology emphasizes applying exclusively deterministic transformations on query expressions to achieve the best possible optimization efficiency. A new approach is thus proposed that facilitates the exploitation of the DTD (document type definition) knowledge, structural properties and structure indices of structured documents for the purpose of fast query optimization.

We present a query system that has been implemented in a practical biological database-GPCRDB. Distinguishing features of this system include: smart query relaxation and smooth integration of navigation with conventional language based query functions. Query relaxation is required due to the fact that queries are not always effective (in other words, expected results are frequently not achieved), particularly in scientific databases like biological databases, in which complex domain knowledge is heavily used. On the other hand, navigation capability is desired as complex data sets are involved, especially in a WWW based environment where multiple hyperlinks are often employed. For efficient implementation, the "fuzzy equivalence class" concept has been applied that captures an important type of domain knowledge.

GPCRDB (G-Protein Coupled Receptors DataBase) is an advanced data management system for G-protein coupled receptors (GPCRs). It collects various data related to all aspects of GPCRs in a variety of forms: HTML pages, formatted source data files, 3D structures, ordinary database tables, etc. The GPCRDB query system is designed to facilitate academic and industrial users in the areas of pharmacology and/or biology to search for information in an integrated, easy-to-operate environment. Currently, the query system is operable on the World Wide Web at.