MapReduce框架下并行知识约简算法模型研究

面向大规模数据进行知识约简是近年来粗糙集理论研究热点。经典的知识约简算法是一次性将小数据集装入单机主存中进行约简,无法处理海量数据。深入剖析了知识约简算法中的可并行性;设计并实现了数据和任务同时并行的Map和Reduce函数,用于计算不同候选属性集导出的等价类和属性重要性;构建了一种MapReduce框架下并行知识约简算法模型,用于计算基于正区域、基于差别矩阵或基于信息熵的知识约简算法的一个约简。在Hadoop平台上进行了相关实验,实验结果表明,该并行知识约简算法模型可以高效地处理海量数据集。     

Top-k dominating queries on incomplete large dataset

Top-k dominating (TKD) query is one of the methods to find the interesting objects by returning the k objects that dominate other objects in a given dataset. Incomplete datasets have missing values in uncertain dimensions, so it is difficult to obtain useful information with traditional data mining methods on complete data. BitMap Index Guided Algorithm (BIG) is a good choice for solving this problem. However, it is even harder to find top-k dominance objects on incomplete big data. When the dataset is too large, the requirements for the feasibility and performance of the algorithm will become very high. In this paper, we proposed an algorithm to apply MapReduce on the whole process with a pruning strategy, called Efficient Hadoop BitMap Index Guided Algorithm (EHBIG). This algorithm can realize TKD query on incomplete datasets through BitMap Index and use MapReduce architecture to make TKD query possible on large datasets. By using the pruning strategy, the runtime and memory usage are greatly reduced. What’s more, we also proposed an improved version of EHBIG (denoted as IEHBIG) which optimizes the whole algorithm flow. Our in-depth work in this article culminates with some experimental results that clearly show that our proposed algorithm can perform well on TKD query in an incomplete large dataset and shows great performance in a Hadoop computing cluster.

     

A MapReduce-based scalable discovery and indexing of structured big data

Various methods and techniques have been proposed in past for improving performance of queries on structured and unstructured data. The paper proposes a parallel B-Tree index in the MapReduce framework for improving efficiency of random reads over the existing approaches. The benefit of using the MapReduce framework is that it encapsulates the complexity of implementing parallelism and fault tolerance from users and presents these in a user friendly way. The proposed index reduces the number of data accesses for range queries and thus improves efficiency. The B-Tree index on MapReduce is implemented in a chained-MapReduce process that reduces intermediate data access time between successive map and reduce functions, and improves efficiency. Finally, five performance metrics have been used to validate the performance of proposed index for range search query in MapReduce, such as, varying cluster size and, size of range search query coverage on execution time, the number of map tasks and size of Input/Output (I/O) data. The effect of varying Hadoop Distributed File System (HDFS) block size and, analysis of the size of heap memory and intermediate data generated during map and reduce functions also shows the superiority of the proposed index. It is observed through experimental results that the parallel B-Tree index along with a chained-MapReduce environment performs better than default non-indexed dataset of the Hadoop and B-Tree like Global Index (Zhao et al., 2012) in MapReduce.     

Generate‐map‐reduce: An extension to map‐reduce to support shared data and recursive computations

It is difficult to express the parallelism present in complex computations by using existing higher level abstractions such as MapReduce and Dryad. These computations include applications from wide variety of domains, like Artificial Intelligence, Decision Tree Algorithms, Association Rule Mining, Recommender Systems, Graph Algorithms, Clustering Algorithms, Compute Intensive Scientific Workflows, Optimization Algorithms, and so forth. Their execution graphs introduce new challenges in terms of programmer expressibility and runtime performance such as iterative and recursive computations, shared communication model, and so forth. We propose an extension to MapReduce, called Generate‐Map‐Reduce (GMR), targeted towards modeling these applications. GMR introduces a new Generate abstraction into the MapReduce framework that captures recursive computations. The runtime also supports iterative jobs and a distributed communication model by using shared data structures. We illustrate recursive computations with GMR by modeling complex applications such as simulated annealing, A* search, and adaptive quadrature computation that require recursive spawning of new tasks to handle variable degree of parallelism. GMR runtime supports caching of common data across iterations in memory and local disks. We illustrate how this caching helps in achieving significant speedup for iterative computations by modeling k‐means clustering. Copyright © 2013 John Wiley & Sons, Ltd.     

Parallel meta-blocking for scaling entity resolution over big heterogeneous data

Entity resolution constitutes a crucial task for many applications, but has an inherently quadratic complexity. In order to enable entity resolution to scale to large volumes of data, blocking is typically employed: it clusters similar entities into (overlapping) blocks so that it suffices to perform comparisons only within each block. To further increase efficiency, Meta-blocking is being used to clean the overlapping blocks from unnecessary comparisons, increasing precision by orders of magnitude at a small cost in recall. Despite its high time efficiency though, using Meta-blocking in practice to solve entity resolution problem on very large datasets is still challenging: applying it to 7.4 million entities takes (almost) 8 full days on a modern high-end server.In this paper, we introduce scalable algorithms for Meta-blocking, exploiting the MapReduce framework. Specifically, we describe a strategy for parallel execution that explicitly targets the core concept of Meta-blocking, the blocking graph. Furthermore, we propose two more advanced strategies, aiming to reduce the overhead of data exchange. The comparison-based strategy creates the blocking graph implicitly, while the entity-based strategy is independent of the blocking graph, employing fewer MapReduce jobs with a more elaborate processing. We also introduce a load balancing algorithm that distributes the computationally intensive workload evenly among the available compute nodes. Our experimental analysis verifies the feasibility and superiority of our advanced strategies, and demonstrates their scalability to very large datasets.     

Analytical Performance Models for MapReduce Workloads

MapReduce is a currently popular programming model to support parallel computations on large datasets. Among the several existing MapReduce implementations, Hadoop has attracted a lot of attention from both industry and research. In a Hadoop job, map and reduce tasks coordinate to produce a solution to the input problem, exhibiting precedence constraints and synchronization delays that are characteristic of a pipeline communication between maps (producers) and reduces (consumers). We here address the challenge of designing analytical models to estimate the performance of MapReduce workloads, notably Hadoop workloads, focusing particularly on the intra-job pipeline parallelism between map and reduce tasks belonging to the same job. We propose a hierarchical model that combines a precedence graph model and a queuing network model to capture the intra-job synchronization constraints. We first show how to build a precedence graph that represents the dependencies among multiple tasks of the same job. We then apply it jointly with an approximate Mean Value Analysis (aMVA) solution to predict mean job response time, throughput and resource utilization. We validate our solution against a queuing network simulator and a real setup in various scenarios, finding very close agreement in both cases. In particular, our model produces estimates of average job response time that deviate from measurements of a real setup by less than 15 %.     

The HaLoop approach to large-scale iterative data analysis

The growing demand for large-scale data mining and data analysis applications has led both industry and academia to design new types of highly scalable data-intensive computing platforms. MapReduce has enjoyed particular success. However, MapReduce lacks built-in support for iterative programs, which arise naturally in many applications including data mining, web ranking, graph analysis, and model fitting. This paper (This is an extended version of the VLDB 2010 paper “HaLoop: Efficient Iterative Data Processing on Large Clusters” PVLDB 3(1):285–296, 2010.) presents HaLoop, a modified version of the Hadoop MapReduce framework, that is designed to serve these applications. HaLoop allows iterative applications to be assembled from existing Hadoop programs without modification, and significantly improves their efficiency by providing inter-iteration caching mechanisms and a loop-aware scheduler to exploit these caches. HaLoop retains the fault-tolerance properties of MapReduce through automatic cache recovery and task re-execution. We evaluated HaLoop on a variety of real applications and real datasets. Compared with Hadoop, on average, HaLoop improved runtimes by a factor of 1.85 and shuffled only 4 % as much data between mappers and reducers in the applications that we tested.     

Fast and scalable vector similarity joins with MapReduce

Vector similarity join, which finds similar pairs of vector objects, is a computationally expensive process. As its number of vectors increases, the time needed for join operation increases proportional to the square of the number of vectors. Various filtering techniques have been proposed to reduce its computational load. On the other hand, MapReduce algorithms have been studied to manage large datasets. The recent improvements, however, still suffer from its computational time and scalability. In this paper, we propose a MapReduce algorithm FACET(FAst and sCalable maprEduce similariTy join) to efficiently solve the vector similarity join problem on large datasets. FACET is an all-pair exact join algorithm, composed of two stages. In the first stage, we use our own novel filtering techniques to eliminate dissimilar pairs to generate non-redundant candidate pairs. The second stage matches candidate pairs with the vector data so that similar pairs are produced as the output. Both stages employ parallelism offered by MapReduce. The algorithm is currently designed for cosine similarity and Self Join case. Extensions to other similarity measures and R-S Join case are also discussed. We provide the I/O analysis of the algorithm. We evaluate the performance of the algorithm on multiple real world datasets. The experiment results show that our algorithm performs, on average, 40 % upto 800 % better than the previous state-of-the-art MapReduce algorithms.     

Improving scalability of inductive logic programming via pruning and best-effort optimisation

Inductive Logic Programming (ILP) combines rule-based and statistical artificial intelligence methods, by learning a hypothesis comprising a set of rules given background knowledge and constraints for the search space. We focus on extending the XHAIL algorithm for ILP which is based on Answer Set Programming and we evaluate our extensions using the Natural Language Processing application of sentence chunking. With respect to processing natural language, ILP can cater for the constant change in how we use language on a daily basis. At the same time, ILP does not require huge amounts of training examples such as other statistical methods and produces interpretable results, that means a set of rules, which can be analysed and tweaked if necessary. As contributions we extend XHAIL with (i) a pruning mechanism within the hypothesis generalisation algorithm which enables learning from larger datasets, (ii) a better usage of modern solver technology using recently developed optimisation methods, and (iii) a time budget that permits the usage of suboptimal results. We evaluate these improvements on the task of sentence chunking using three datasets from a recent SemEval competition. Results show that our improvements allow for learning on bigger datasets with results that are of similar quality to state-of-the-art systems on the same task. Moreover, we compare the hypotheses obtained on datasets to gain insights on the structure of each dataset.     

HFIM: a Spark-based hybrid frequent itemset mining algorithm for big data processing

Frequent itemset mining is one of the data mining techniques applied to discover frequent patterns, used in prediction, association rule mining, classification, etc. Apriori algorithm is an iterative algorithm, which is used to find frequent itemsets from transactional dataset. It scans complete dataset in each iteration to generate the large frequent itemsets of different cardinality, which seems better for small data but not feasible for big data. The MapReduce framework provides the distributed environment to run the Apriori on big transactional data. However, MapReduce is not suitable for iterative process and declines the performance. We introduce a novel algorithm named Hybrid Frequent Itemset Mining (HFIM), which utilizes the vertical layout of dataset to solve the problem of scanning the dataset in each iteration. Vertical dataset carries information to find support of each itemsets. Moreover, we also include some enhancements to reduce number of candidate itemsets. The proposed algorithm is implemented over Spark framework, which incorporates the concept of resilient distributed datasets and performs in-memory processing to optimize the execution time of operation. We compare the performance of HFIM with another Spark-based implementation of Apriori algorithm for various datasets. Experimental results show that the HFIM performs better in terms of execution time and space consumption.     

Efficient and exact duplicate detection on cloud

As the recent proliferation of social networks, mobile applications, and online services increased the rate of data gathering, to find near‐duplicate records efficiently has become a challenging issue. Related works on this problem mainly aim to propose efficient approaches on a single machine. However, when processing large‐scale dataset, the performance to identify duplicates is still far from satisfactory. In this paper, we try to handle the problem of duplicate detection applying MapReduce. We argue that the performance of utilizing MapReduce to detect duplicates mainly depends on the number of candidate record pairs and intermediate result size, which is related to the shuffle cost among different nodes in cluster. In this paper, we proposed a new signature scheme with new pruning strategies to minimize the number of candidate pairs and intermediate result size. The proposed solution is an exact one, which assures none duplicate record pair can be lost. The experimental results over both real and synthetic datasets demonstrate that our proposed signature‐based method is efficient and scalable. Copyright © 2012 John Wiley & Sons, Ltd.     

Scalable and fast SVM regression using modern hardware

Support Vector Machine (SVM) regression is an important technique in data mining. The SVM training is expensive and its cost is dominated by: (i) the kernel value computation, and (ii) a search operation which finds extreme training data points for adjusting the regression function in every training iteration. Existing training algorithms for SVM regression are not scalable to large datasets because: (i) each training iteration repeatedly performs expensive kernel value computations, which is inefficient and requires holding the whole training dataset in memory; (ii) the search operation used in each training iteration considers the whole search space which is very expensive. In this article, we significantly improve the scalability and efficiency of SVM regression by exploiting the high performance of Graphics Processing Units (GPUs) and solid state drives (SSDs). Our key ideas are as follows. (i) To reduce the cost of repeated kernel value computations and avoid holding the whole training dataset in the GPU memory, we precompute all the kernel values and store them in the CPU memory extended by the SSD; together with an efficient strategy to read the precomputed kernel values, reusing precomputed kernel values with an efficient retrieval is much faster than computing them on-the-fly. This also alleviates the restriction that the training dataset has to fit into the GPU memory, and hence makes our algorithm scalable to large datasets, especially for large datasets with very high dimensionality. (ii) To enhance the performance of the frequently used search operation, we design an algorithm that minimizes the search space and the number of accesses to the GPU global memory; this optimized search algorithm also avoids branch divergence (one of the causes for poor performance) among GPU threads to achieve high utilization of the GPU resources. Our proposed techniques together form a scalable solution to the SVM regression which we call SIGMA. Our extensive experimental results show that SIGMA is highly efficient and can handle very large datasets which the state-of-the-art GPU-based algorithm cannot handle. On the datasets of size that the state-of-the-art GPU-based algorithm can handle, SIGMA consistently outperforms the state-of-the-art GPU-based algorithm by an order of magnitude and achieves up to 86 times speedup.     

Master–worker model for MapReduce paradigm on the TILE64 many-core platform

MapReduce is a popular programming paradigm for processing big data. It uses the master–worker model, which is widely used on distributed and loosely coupled systems such as clusters, to solve large problems with task parallelism. With the ubiquity of many-core architectures in recent years and foreseeable future, the many-core platform will be one of the main computing platforms to execute MapReduce programs. Therefore, it is essential to optimize MapReduce programs on many-core platforms. Optimizations of parallel programs for a many-core platform are viewed as a multifaceted problem, where both system and architectural factors should be taken into account. In this paper, we look into the problem by constructing a master–worker model for MapReduce paradigm on the TILE64 many-core platform. We investigate master share and worker share schemes for implementation of a MapReduce library on the TILE64. The theoretical analysis shows that the worker share scheme is inherently better for implementation of MapReduce library on the TILE64 many-core platform.     

Compilation techniques for parallel systems

Over the past two decades tremendous progress has been made in both the design of parallel architectures and the compilers needed for exploiting parallelism on such architectures. In this paper we summarize the advances in compilation techniques for uncovering and effectively exploiting parallelism at various levels of granularity. We begin by describing the program analysis techniques through which parallelism is detected and expressed in form of a program representation. Next compilation techniques for scheduling instruction level parallelism (ILP) are discussed along with the relationship between the nature of compiler support and type of processor architecture. Compilation techniques for exploiting loop and task level parallelism on shared-memory multiprocessors (SMPs) are summarized. Locality optimizations that must be used in conjunction with parallelization techniques for achieving high performance on machines with complex memory hierarchies are also discussed. Finally we provide an overview of compilation techniques for distributed memory machines that must perform partitioning of both code and data for parallel execution. Communication optimization and code generation issues that are unique to such compilers are also briefly discussed.     

Task Selection for the Multiscalar Architecture

The multiscalar architecture advocates a distributed processor organization and task-level speculation to exploit high degrees of instruction level parallelism (ILP) in sequential programs without impeding improvements in clock speeds. The main goal of this paper is to understand the key implications of the architectural features of distributed processor organization and task-level speculation for compiler task selection from the point of view of performance. We identify the fundamental performance issues to be: control flow speculation, data communication, data dependence speculation, load imbalance, and task overhead. We show that these issues are intimately related to a few key characteristics of tasks: task size, intertask control flow, and intertask data dependence. We describe compiler heuristics to select tasks with favorable characteristics. We report experimental results to show that the heuristics are successful in boosting overall performance by establishing larger ILP windows. We also present a breakdown of execution times to show that register wait, load imbalance, control flow squash, and conventional pipeline losses are significant for almost all the SPEC95 benchmarks.     

Parallel construction of classification trees on a GPU

Algorithms for constructing tree‐based classifiers are aimed at building an optimal set of rules implicitly described by some dataset of training samples. As the number of samples and/or attributes in the dataset increases, the required construction time becomes the limiting factor for interactive or even functional use. The problem is emphasized if tree derivation is part of an iterative optimization method, such as boosting. Attempts to parallelize the construction of classification trees have therefore been presented in the past. This paper describes a parallel method for binary classification tree construction implemented on a graphics processing unit (GPU) using compute unified device architecture (CUDA). By employing node‐level, attribute‐level, and split‐level parallelism, the task parallel and data parallel sections of tree induction are mapped to the architecture of a modern GPU. The GPU‐based solution is compared with the sequential and multi‐threaded CPU versions on public access datasets, and it is shown that an order of magnitude acceleration can be achieved on this data‐intensive task using inexpensive commodity hardware. The influence of dataset characteristics on the efficiency of parallel implementation is also analyzed. Copyright © 2015 John Wiley & Sons, Ltd.     

Relational large scale multi-label classification method for video categorization

The problem of automated video categorization in large datasets is considered in the paper. A new Iterative Multi-label Propagation (IMP) algorithm for relational learning in multi-label data is proposed. Based on the information of the already categorized videos and their relations to other videos, the system assigns suitable categories—multiple labels to the unknown videos. The MapReduce approach to the IMP algorithm described in the paper enables processing of large datasets in parallel computing. The experiments carried out on 5-million videos dataset revealed the good efficiency of the multi-label classification for videos categorization. They have additionally shown that classification of all unknown videos required only several parallel iterations.     

基于Relief和BFO的并行支持向量机算法

针对大数据环境下并行支持向量机(SVM)算法存在冗余数据敏感、参数选取困难、并行化效率低等问题,提出了一种基于Relief和BFO算法的并行SVM算法RBFO-PSVM。首先,基于互信息和Relief算法设计了一种特征权值计算策略MI-Relief,剔除数据集中的冗余特征,有效地降低了冗余数据对并行SVM分类的干扰;接着,提出了基于MapReduce的MR-HBFO算法,并行选取SVM的最优参数,提高SVM的参数寻优能力;最后,提出核聚类策略KCS,减小参与并行化训练的数据集规模,并提出改进CSVM反馈机制的交叉融合级联式并行支持向量机CFCPSVM,结合MapReduce编程框架并行训练SVM,提高了并行SVM的并行化效率。实验表明,RBFO-PSVM算法对大型数据集的分类效果更佳,更适用于大数据环境。     

In-memory,distributed content-based recommender system

Burdened by their popularity, recommender systems increasingly take on larger datasets while they are expected to deliver high quality results within reasonable time. To meet these ever growing requirements, industrial recommender systems often turn to parallel hardware and distributed computing. While the MapReduce paradigm is generally accepted for massive parallel data processing, it often entails complex algorithm reorganization and suboptimal efficiency because mid-computation values are typically read from and written to hard disk. This work implements an in-memory, content-based recommendation algorithm and shows how it can be parallelized and efficiently distributed across many homogeneous machines in a distributed-memory environment. By focusing on data parallelism and carefully constructing the definition of work in the context of recommender systems, we are able to partition the complete calculation process into any number of independent and equally sized jobs. An empirically validated performance model is developed to predict parallel speedup and promises high efficiencies for realistic hardware configurations. For the MovieLens 10 M dataset we note efficiency values up to 71 % for a configuration of 200 computing nodes (eight cores per node).