Lars Kolb

Load Balancing for MapReduce-based Entity Resolution

The effectiveness and scalability of MapReduce-based implementations of complex data-intensive tasks depend on an even redistribution of data between map and reduce tasks. In the presence of skewed data, sophisticated redistribution approaches thus become necessary to achieve load balancing among all reduce tasks to be executed in parallel. For the complex problem of entity resolution, we propose and evaluate two approaches for such skew handling and load balancing. The approaches support blocking techniques to reduce the search space of entity resolution, utilize a preprocessing MapReduce job to analyze the data distribution, and distribute the entities of large blocks among multiple reduce tasks. The evaluation on a real cloud infrastructure shows the value and effectiveness of the proposed load balancing approaches.

Dedoop: efficient deduplication with Hadoop

We demonstrate a powerful and easy-to-use tool called Dedoop ( De duplication with Ha doop ) for MapReduce-based entity resolution (ER) of large datasets. Dedoop supports a browser-based specification of complex ER workflows including blocking and matching steps as well as the optional use of machine learning for the automatic generation of match classifiers. Specified workflows are automatically translated into MapReduce jobs for parallel execution on different Hadoop clusters. To achieve high performance Dedoop supports several advanced load balancing strategies.

Multi-pass sorted neighborhood blocking with MapReduce

Cloud infrastructures enable the efficient parallel execution of data-intensive tasks such as entity resolution on large datasets. We investigate challenges and possible solutions of using the MapReduce programming model for parallel entity resolution using Sorting Neighborhood blocking (SN). We propose and evaluate two efficient MapReduce-based implementations for single- and multi-pass SN that either use multiple MapReduce jobs or apply a tailored data replication. We also propose an automatic data partitioning approach for multi-pass SN to achieve load balancing. Our evaluation based on real-world datasets shows the high efficiency and effectiveness of the proposed approaches.

Parallel Entity Resolution with Dedoop

We provide an overview of Dedoop (Deduplication with Hadoop), a new tool for parallel entity resolution (ER) on cloud infrastructures. Dedoop supports a browser-based specification of complex ER strategies and provides a large library of blocking and matching approaches. To simplify the configuration of ER strategies with several similarity metrics, training-based machine learning approaches can be employed with Dedoop. Specified ER strategies are automatically translated into MapReduce jobs for parallel execution on different Hadoop clusters. For improved performance, Dedoop supports redundancy-free multi-pass blocking as well as advanced load balancing approaches. To illustrate the usefulness of Dedoop, we present the results of a comparative evaluation of different ER strategies on a challenging real-world dataset.

When to Reach for the Cloud: Using Parallel Hardware for Link Discovery

With the ever-growing amount of RDF data available across the Web, the discovery of links between datasets and deduplication of resources within knowledge bases have become tasks of crucial importance. Over the last years, several link discovery approaches have been developed to tackle the runtime and complexity problems that are intrinsic to link discovery. Yet, so far, little attention has been paid to the management of hardware resources for the execution of link discovery tasks. This paper addresses this research gap by investigating the efficient use of hardware resources for link discovery. We implement the \(\mathcal{HR}^3\) approach for three different parallel processing paradigms including the use of GPUs and MapReduce platforms. We also perform a thorough performance comparison for these implementations. Our results show that certain tasks that appear to require cloud computing techniques can actually be accomplished using standard parallel hardware. Moreover, our evaluation provides break-even points that can serve as guidelines for deciding on when to use which hardware for link discovery.

Don't match twice: redundancy-free similarity computation with MapReduce

To improve the effectiveness of pair-wise similarity computation, state-of-the-art approaches assign objects to multiple overlapping clusters. This introduces redundant pair comparisons when similar objects share more than one cluster. We propose an approach that eliminates such redundant comparisons and that can be easily integrated into existing MapReduce implementations. We evaluate the approach on a real cloud infrastructure and show its effectiveness for all degrees of redundancy.

Block-based load balancing for entity resolution with MapReduce

The effectiveness and scalability of MapReduce-based implementations of complex data-intensive tasks depend on an even redistribution of data between map and reduce tasks. In the presence of skewed data, sophisticated redistribution approaches thus become necessary to achieve load balancing among all reduce tasks to be executed in parallel. For the complex problem of entity resolution with blocking, we propose BlockSplit, a load balancing approach that supports blocking techniques to reduce the search space of entity resolution. The evaluation on a real cloud infrastructure shows the value and effectiveness of the proposed approach.

Optimizing Similarity Computations for Ontology Matching - Experiences from GOMMA

An efficient computation of ontology mappings requires optimized algorithms and significant computing resources especially for large life science ontologies. We describe how we optimized n-gram matching for computing the similarity of concept names and synonyms in our match system GOMMA. Furthermore, we outline how to enable a highly parallel string matching on Graphical Processing Units (GPU). The evaluation on the OAEI LargeBio match task demonstrates the high effectiveness of the proposed optimizations and that the use of GPUs in addition to standard processors enables significant performance improvements.

Learning-based entity resolution with MapReduce

Entity resolution is a crucial step for data quality and data integration. Learning-based approaches show high effectiveness at the expense of poor efficiency. To reduce the typically high execution times, we investigate how learning-based entity resolution can be realized in a cloud infrastructure using MapReduce. We propose and evaluate two efficient MapReduce-based strategies for pair-wise similarity computation and classifier application on the Cartesian product of two input sources. Our evaluation is based on real-world datasets and shows the high efficiency and effectiveness of the proposed approaches.

Iterative Computation of Connected Graph Components with MapReduce

The use of the MapReduce framework for iterative graph algorithms is challenging. To achieve high performance it is critical to limit the amount of intermediate results as well as the number of necessary iterations. We address these issues for the important problem of finding connected components in large graphs. We analyze an existing MapReduce algorithm, CC-MR, and present techniques to improve its performance including a memory-based connection of subgraphs in the map phase. Our evaluation with several large graph datasets shows that the improvements can substantially reduce the amount of generated data by up to a factor of 8.8 and runtime by up to factor of 3.5.