Band 58, Heft 4 - Big Data Analytics / Guest Editor: Erhard Rahm

Recent trends in data collection and the decreasing prices of storage result in constantly growing amounts of analyzable data. These masses of data cannot easily be processed by traditional database systems as these do not allow for a sufficient degree of scalability. Programs especially designed for parallel data analysis on large-scale distributed systems are required. Developing such programs on clusters of commodity hardware is a complex challenge for even the most experienced system developers. Frameworks such as Apache Hadoop are scalable, but – when compared to SQL – extremely hard to program. The open-source platform Apache Flink is a link between conventional database systems and big data analysis frameworks. Flink is based on a fault tolerant runtime for data stream processing, which manages the distribution of data as well as communications within the cluster. A high diversity of use cases can be supported through various interfaces that allow for the implementation of data analysis processes. In this paper, we present an overview of Apache Flink as well as some current research activities on top of the Apache Flink ecosystem.

Using graph data models for business intelligence applications is a novel and promising approach. In contrast to traditional data warehouse models, graph models enable the mining of relationship patterns. In our prior work, we introduced an approach to graph-based data integration and analytics called BIIIG (Business Intelligence with Integrated Instance Graphs). In this work, we compare state-of-the-art systems for graph data management and analytics with regard to the support for our approach in Big Data scenarios. To exemplify the analytical value of graph models for business intelligence, we propose an analytical workflow to extract knowledge from graph-integrated business data. Finally, we show how we use Gradoop, a novel framework for distributed graph analytics, to implement our approach.

Big Data and Big Data analytics have attracted major interest in research and industry and continue to do so. The high demand for capable and scalable analytics in combination with the ever increasing number and volume of application scenarios and data has lead to a large and intransparent landscape full of versions, variants and individual algorithms. As this zoo of methods lacks a systematic way of description, understanding is almost impossible which severely hinders effective application and efficient development of analytic algorithms. To solve this issue we propose our concept of modular analytics that abstracts the essentials of an analytic domain and turns them into a set of universal building blocks. As arbitrary algorithms can be created from the same set of blocks, understanding is eased and development benefits from reusability.

With the rise of the web 2.0 and the Internet of things, it has become feasible to track all kinds of information over time, in particular fine-grained user activities and sensor data on their environment and even their biometrics. However, while efficiency remains mandatory for any application trying to cope with huge amounts of data, only part of the potential of today's Big Data repositories can be exploited using traditional batch-oriented approaches as the value of data often decays quickly and high latency becomes unacceptable in some applications. In the last couple of years, several distributed data processing systems have emerged that deviate from the batch-oriented approach and tackle data items as they arrive, thus acknowledging the growing importance of timeliness and velocity in Big Data analytics. In this article, we give an overview over the state of the art of stream processors for low-latency Big Data analytics and conduct a qualitative comparison of the most popular contenders, namely Storm and its abstraction layer Trident, Samza and Spark Streaming. We describe their respective underlying rationales, the guarantees they provide and discuss the trade-offs that come with selecting one of them for a particular task.

Nowadays, data is produced in every aspect of our lives, leading to a massive amount of information generated every second. However, this vast amount is often too large to be stored and for many applications the information contained in these data streams is only useful when it is fresh. Batch processing platforms like Hadoop MapReduce do not fit these needs as they require to collect data on disk and process it repeatedly. Therefore, modern data processing engines combine the scalability of distributed architectures with the one-pass semantics of traditional stream engines. In this paper, we survey the current state of the art in scalable stream processing from a user perspective. We examine and describe their architecture, execution model, programming interface, and data analysis support as well as discuss the challenges and limitations of their APIs. In this connection, we introduce Piglet, an extended Pig Latin language and code generator that compiles (extended) Pig Latin code into programs for various data processing platforms. Thereby, we discuss the mapping to platform-specic concepts in order to provide a uniform view.

Traditional clustering algorithms merely considered static data. Today's various applications and research issues in big data mining have however to deal with continuous, possibly infinite streams of data, arriving at high velocity . Web traffic data, surveillance data, sensor measurements and stock trading are only some examples of these daily-increasing applications. Since the growth of data volumes is accompanied by a similar raise in their dimensionalities, clusters cannot be expected to completely appear when considering all attributes together. Subspace clustering is a general approach that solved that issue by automatically finding the hidden clusters within different subsets of the attributes rather than considering all attributes together. In this article, novel methods for an efficient subspace clustering of high-dimensional big data streams are presented. Approaches that efficiently combine the anytime clustering concept with the stream subspace clustering paradigm are discussed. Additionally, efficient and adaptive density-based clustering algorithms are presented for high-dimensional data streams. Novel open-source assessment framework and evaluation measures are additionally presented for subspace stream clustering.