Finding the optimal configuration of a black-box system is a difficult problem that requires a lot of time and human labor. Big data processing frameworks are among the increasingly popular systems whose tuning is a complex and time consuming. The challenge of automatically finding the optimal parameters of big data frameworks attracted a lot of research in recent years. Some of the studies focused on optimizing specific frameworks such as distributed stream processing [1], [2], or finding the best cloud configurations [3], while others proposed general services for optimizing any black-box system [4]. In this paper, we introduce a new use case in the domain of automatic parameter tuning: optimizing the parameters of distributed graph processing frameworks. This task is notably difficult given the particular challenges of distributed graph processing that include the graph partitioning and the iterative nature of graph algorithms. To address this challenge, we designed and implemented GraphOpt: an efficient and scalable black-box optimization framework that automatically tunes distributed graph processing frameworks. GraphOpt implements state-of-the-art optimization algorithms and introduces a new hill-climbing-based search algorithm. These algorithms are used to optimize the performance of two major graph processing frameworks: Giraph and GraphX. Extensive experiments were run on GraphOpt using multiple graph benchmarks to evaluate its performance and show that it provides up to 47.8% improvement compared to random search and an average improvement of up to 5.7%.

Due to the increasing importance and volume of highly interconnected data, such as in social or information networks, a plethora of graph mining techniques have been designed to enable the analysis of such data. In this work, we focus on the mining of associations between entity features in networks. We model each entity feature as a dimension to be analyzed. Consequently we build our approach on top of the existing graph cube framework which is an extension of the concept of the data cube to networks. Our task is particularly challenging because it requires the analysis of both the initial multidimensional network and all its subsequent aggregate forms. As soon as we deal with a big data situation it is impossible for an analyst to consider manually all the possible views of the network data. The aim of this work is to design an algorithm for the discovery of interesting patterns in large graph cubes. Thus, instead of examining all the possible aggregations manually, the proposed technique leads the analyst to the interesting associations or patterns in the multidimensional network. Furthermore, we study the application of existing algorithms from the frequent itemset mining literature on graph data and propose a mapping between the two settings.

Iterative-convergent algorithms represent an important family of applications in big data analytics. These are typically run on distributed processing frameworks deployed on a cluster of machines. On the other hand, we are witnessing the move towards data center operating systems (OS), where resources are unified by a resource manager and processing frameworks coexist with each other. In this context, different processing framework job tasks can be scheduled on the same machine and slow down a worker (straggler problem). Existing work has shown that an iteration model with relaxed consistency such as the Stale Synchronous Parallel (SSP) model, while still guaranteeing convergence, is able to cope with stragglers. In this paper we propose a model for the integration of the SSP model on a pipelined distributed processing framework. We then apply SSP on a distributed version of the Frank-Wolfe algorithm. We theoretically show its sparsity bounds and convergence under SSP. Finally, we experimentally show that the Frank-Wolfe algorithm applied on LASSO regression under SSP is able to converge faster than its BSP counterpart, especially under load conditions similar to those encountered in a data center OS.

The importance of graphs as the fundamental structure underpinning many real world applications is no longer to be proved. Large graphs have emerged in various fields such as biological, social and transportation networks. The sheer volume of these networks poses challenges to traditional techniques for storage and analysis of graph data. In particular, OLAP analysis requires access to large portions of data to extract key information and to feed strategic decision making. OLAP provides multilevel, multiperspective views of the data. Most of the current techniques are optimized for centralized graph processing. A distributed approach providing horizontal scalability is required in order to handle the analysis workload. In this paper, we focus on applying OLAP analysis on large, distributed graph data. We describe Distributed Graph Cube, our distributed framework for graph-based OLAP cubes computation and aggregation. Experimental results on large, real-world datasets demonstrate that our method significantly outperforms its centralized counterparts. We also evaluate the performance of both Hadoop and Spark for distributed cubes computations.