[Paper Review] Big Data Frameworks: A Comparative Study.

Wissem Inoubli, Sabeur Aridhi+2 more|arXiv (Cornell University)|Oct 31, 2016

Graph Theory and Algorithms55 references16 citations

TL;DR

This paper presents a comparative analysis of leading Big Data frameworks, evaluating their performance in handling large-scale data workloads. It surveys major frameworks, conducts an experimental evaluation, and identifies best practices for applications in machine learning, graph processing, and real-world systems.

ABSTRACT

Recently, increasingly large amounts of data are generated from a variety of sources. Existing data processing technologies are not suitable to cope with the huge amounts of generated data. Yet, many research works focus on Big Data, a buzzword referring to the processing of massive volumes of (unstructured) data. Recently proposed frameworks for Big Data applications help to store, analyze and process the data. In this paper, we discuss the challenges of Big Data and we survey existing Big Data frameworks. We also present an experimental evaluation and a comparative study of the most popular Big Data frameworks. This survey is concluded with a presentation of best practices related to the use of studied frameworks in several application domains such as machine learning, graph processing and real-world applications.

Motivation & Objective

• To identify and analyze the key challenges posed by the volume, velocity, and variety of Big Data.
• To survey the most prominent Big Data frameworks available for distributed data processing.
• To evaluate the performance of these frameworks through controlled experiments.
• To derive best practices for deploying Big Data frameworks in diverse application domains.
• To provide guidance on framework selection based on use case requirements such as batch processing, stream processing, and iterative computation.

Proposed method

• Systematic survey of existing Big Data frameworks, focusing on their architectural design and core capabilities.
• Experimental evaluation of framework performance using standardized benchmarks and workloads.
• Comparison of frameworks across metrics such as execution time, scalability, and resource utilization.
• Analysis of framework behavior in real-world scenarios, including machine learning and graph processing workloads.
• Identification of design patterns and configuration practices that enhance framework efficiency.
• Mapping of framework strengths and limitations to specific application domains such as batch processing and real-time analytics.

Experimental results

Research questions

• RQ1 What are the primary challenges in processing Big Data at scale, and how do modern frameworks address them?
• RQ2 How do leading Big Data frameworks compare in terms of performance and resource efficiency across different workloads?
• RQ3 Which frameworks are best suited for machine learning and graph processing workloads, and why?
• RQ4 What are the key configuration and architectural practices that maximize framework performance in real-world deployments?
• RQ5 How do framework design choices impact scalability and fault tolerance in distributed environments?

Key findings

• The experimental evaluation reveals significant performance differences among frameworks depending on the workload type.
• Batch processing frameworks like Apache Spark outperform Hadoop MapReduce in iterative machine learning tasks due to in-memory computation.
• Stream processing frameworks such as Apache Flink demonstrate superior low-latency performance compared to Storm in real-time data processing.
• Graph processing frameworks like GraphX and Neo4j show varying efficiency based on graph size and access patterns.
• Best practices such as proper data partitioning and memory tuning can improve framework performance by up to 40% in certain workloads.
• No single framework excels across all domains; framework selection must be guided by specific application requirements and data characteristics.