Scholarlink

<p>Graph processing frameworks are pivotal in managing the increasing complexity and size of datasets in numerous computational fields. However, the performance of these frameworks can vary significantly depending on their design and the specific conditions under which they are tested. This paper presents a comprehensive benchmarking study of six single-machine graph processing frameworks: GraphChi, Ligra, X-stream, MMap, GridGraph, and GPOP across a variety of datasets and graph processing algorithms, and on a consumer-level machine. In the evaluation of the frameworks, several performance metrics were used, which are execution time, CPU, and memory usage, as well as qualitative metrics, such as ease of use and comprehensive documentation which contribute to determining each framework’s practical effectiveness. Our findings indicate that no single framework excels in all aspects; instead, each framework demonstrates strengths and weaknesses that are context-dependent. By evaluating frameworks in a unified environment rather than their original setups, the study explores which frameworks perform more effectively under comparable conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Scaling down to scale up: benchmarking single-machine graph processing frameworks in a hardware-constrained environment

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Mariem Loukil,
  • Lilia Sfaxi,
  • Riadh Robbana

摘要

Graph processing frameworks are pivotal in managing the increasing complexity and size of datasets in numerous computational fields. However, the performance of these frameworks can vary significantly depending on their design and the specific conditions under which they are tested. This paper presents a comprehensive benchmarking study of six single-machine graph processing frameworks: GraphChi, Ligra, X-stream, MMap, GridGraph, and GPOP across a variety of datasets and graph processing algorithms, and on a consumer-level machine. In the evaluation of the frameworks, several performance metrics were used, which are execution time, CPU, and memory usage, as well as qualitative metrics, such as ease of use and comprehensive documentation which contribute to determining each framework’s practical effectiveness. Our findings indicate that no single framework excels in all aspects; instead, each framework demonstrates strengths and weaknesses that are context-dependent. By evaluating frameworks in a unified environment rather than their original setups, the study explores which frameworks perform more effectively under comparable conditions.