A Survey on Data-driven Performance Tuning for Big Data Analytics Platforms

Abstract

Many research works deal with big data platforms looking forward to data science and analytics. These are complex and usually distributed environments, composed of several systems and tools. As expected, there is a need for a closer look at performance issues.

In this work, we review performance tuning strategies in the big data environment. We focus on data-driven tuning techniques, discussing the use of database inspired approaches. Concerning big data and NoSQL stores, performance tuning issues are quite different from the so-called conventional systems. Many existing solutions are mostly ad-hoc activities that do not fit for multiple situations. But there are some categories of data-driven solutions that can be taken as guidelines and incorporated into general-purpose auto-tuning modules for big data systems.

We examine typical performance tuning actions, discussing available solutions to support some of the tuning process's primary activities. We also discuss recent implementations of data-driven performance tuning solutions for big data platforms. We propose an initial classification based on the domain state-of-the-art and present selected tuning actions for large-scale data processing systems. Finally, we organized existing works towards self-tuning big data systems based on this classification and presented general and system-specific tuning recommendations. We found that most of the literature pieces evaluate the use of tuning actions at the physical design perspective, and there is a lack of self-tuning machine-learning-based solutions for big data systems.

Introduction

A wide range of devices, including mobile phones, GPS devices, social networks, sensors, and IoT devices [1], [2], [3], [4] is generating a large volume of distributed and heterogeneous data. Transforming such massive amount of data into valuable information while revealing its underlying meaning is a crucial function of big data analytics [5], [6].

New requirements in terms of analytics (e.g., real-time data management) contribute to a big data analytics environment composed of a myriad of solutions and distributed systems, commonly known as the big data stack [5]. Many of these solutions involve distributed processing of large volumes of data, replicated or moved over networks on the fly. Failures are somewhat common, and runtime auto-recovery is required and commonly achieved based on re-processing of tasks and through the use of intermediate data materialization [7], [8]. In such an environment, I/O costs and data movement may impose relevant costs and overheads. Capacity planning may turn out to be a problem and lead to increasing costs, especially in applications that must meet some type of Service Level Agreement (SLA) specified in terms of performance metrics [5]. Also, there is no optimizer to build the most efficient execution plan to execute each user's job, which increases the importance of carefully tuning the system performance.

Both automatic tuning and real-time analytics tuning are some of the main open challenges for big data systems [9]. In big data analytics environments, the main tuning opportunities include the reorganization of data layouts (e.g., applying partitioning) and the efficient use of materialization, mainly to persist intermediate results [10]. These opportunities are somewhat related to some of the most common performance tuning actions usually taken in traditional database systems, which are data-driven decisions, mostly oriented to reduce I/O operations and improve memory consumption. Some of the aspects that impact traditional database systems' performance also affect large-scale big data processing systems. For instance, query (job) size, data skew, and temporal locality in data access impact both database systems and MapReduce-based systems performance [11].

In this paper, we review the use of performance tuning techniques in the big data stack. We focus on data-driven performance tuning, discussing works related to I/O optimization, data partitioning and materialization, file placement and data transfer, caching, and memory consumption. We describe traditional tuning activities and present how it has evolved into the NoSQL solutions. We describe proposals on data-driven performance tuning for big data systems and warehouses, mostly in solutions based on the MapReduce paradigm and distributed systems. We present selected tuning actions that would benefit data analysts. We also present big data benchmarking tools, self-tuning solutions, and performance monitoring tools and advisors, and discuss some future research opportunities.

For the best of our knowledge, there is no prior work that presents an organized review of tuning techniques and works in the context of big data systems and analytics with such a data-driven perspective.

In the following section, we describe some background on the big data systems. We also discuss some related big data benchmarks and studies on performance comparison. Section 3 describes the main performance tuning actions, discussing available solutions used to support some of the tuning process's main activities in relational and NoSQL stores, based on our data-driven performance tuning classification. Section 4 reviews data-driven performance (self)tuning in the large-scale processing systems. In Section 5, we classify existing works, extract general and system-specific tuning recommendations, and discuss research opportunities. Finally, section 6 concludes the paper and presents future work.