Scholarlink

<p>The growing demand for data-intensive parallel applications in cloud computing requires a cost-effective approach to utilizing distributed processing power. However, due to the various instance types and configurations available, choosing the right instance type and configuration to execute any given workload is essential to maximize workload performance while minimizing costs. In this context, our comprehensive evaluation of 104 instances across three major cloud providers for eighteen parallel workloads reveals that no single instance excels in all metrics. While high-performance computing (HPC) optimized instances can improve performance by up to 66% for scaling workloads, they are not always cost-effective. In contrast, general-purpose instances provide a better performance-cost balance for less demanding tasks. Therefore, we show that selecting the most suitable instance based on the workload characteristics can lead to a 35.7% performance improvement and a 6.5% cost reduction compared to using only one instance type for all workloads.</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

Investigating cloud instances to achieve optimal trade-offs between performance-cost efficiency

  • 期刊论文

  • 发表日期：

    2025年2月27日

查看全文
  • William Maas,
  • Fábio Rossi,
  • Marcelo Caggiani,
  • Philippe Navaux,
  • Arthur Lorenzon

摘要

The growing demand for data-intensive parallel applications in cloud computing requires a cost-effective approach to utilizing distributed processing power. However, due to the various instance types and configurations available, choosing the right instance type and configuration to execute any given workload is essential to maximize workload performance while minimizing costs. In this context, our comprehensive evaluation of 104 instances across three major cloud providers for eighteen parallel workloads reveals that no single instance excels in all metrics. While high-performance computing (HPC) optimized instances can improve performance by up to 66% for scaling workloads, they are not always cost-effective. In contrast, general-purpose instances provide a better performance-cost balance for less demanding tasks. Therefore, we show that selecting the most suitable instance based on the workload characteristics can lead to a 35.7% performance improvement and a 6.5% cost reduction compared to using only one instance type for all workloads.