<p>Cloud computing enables on-demand resource allocation, but selecting the appropriate resources for efficient auto-scaling remains a significant challenge. Traditional auto-scaling approaches, which rely on a static type of resources to scale strategies, often result in inefficient resource utilization and increased operational costs. One important but often overlooked factor in auto-scaling decisions is the selection of the optimal virtual machine (VM) type, which directly affects the system performance and cost efficiency. This paper presents a dynamic auto-scaling method that optimally selects VM types based on workload demands. By leveraging Reinforcement Learning, our approach identifies the precise resource requirements and dynamically provisions the most suitable VM type to maintain performance while minimizing the overall costs. We conduct extensive evaluations and compare our method with different auto-scaling techniques using real-time workloads. The experimental results show that the proposed auto-scaling method significantly outperforms baseline approaches. The proposed method handles on average 10.49% more process requests, with 7.89<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="607_2025_1535_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">\(\times \)</EquationSource> </InlineEquation> fewer SLO violations as compared to the baseline methods for five different real-time workloads, including ClarkNet, NASA, Wikipedia, Calgary, and the FIFA 1998 World Cup. Moreover, the proposed methods reduce the 1.37<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="607_2025_1535_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">\(\times \)</EquationSource> </InlineEquation> resource utilization cost as compared to baseline methods.</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

Cost and performance-effective dynamic VM type selection in auto-scaling using reinforcement learning

  • Muhammad Abdullah,
  • Rida Munir,
  • Waheed Iqbal,
  • Saad Ahmad Khan,
  • Shuja Ur Rehman Baig,
  • Abdelkarim Erradi

摘要

Cloud computing enables on-demand resource allocation, but selecting the appropriate resources for efficient auto-scaling remains a significant challenge. Traditional auto-scaling approaches, which rely on a static type of resources to scale strategies, often result in inefficient resource utilization and increased operational costs. One important but often overlooked factor in auto-scaling decisions is the selection of the optimal virtual machine (VM) type, which directly affects the system performance and cost efficiency. This paper presents a dynamic auto-scaling method that optimally selects VM types based on workload demands. By leveraging Reinforcement Learning, our approach identifies the precise resource requirements and dynamically provisions the most suitable VM type to maintain performance while minimizing the overall costs. We conduct extensive evaluations and compare our method with different auto-scaling techniques using real-time workloads. The experimental results show that the proposed auto-scaling method significantly outperforms baseline approaches. The proposed method handles on average 10.49% more process requests, with 7.89 \(\times \) fewer SLO violations as compared to the baseline methods for five different real-time workloads, including ClarkNet, NASA, Wikipedia, Calgary, and the FIFA 1998 World Cup. Moreover, the proposed methods reduce the 1.37 \(\times \) resource utilization cost as compared to baseline methods.