Scholarlink

<p>The cold start latency issue in serverless computing remains a significant challenge, especially for large or dependency-heavy functions. This study introduces ServerlessPGO, which applies Profile-Guided Optimization (PGO) techniques to serverless architectures. ServerlessPGO analyzes runtime execution patterns to optimize dependencies and code paths, reducing function initialization overhead. Our implementation focuses on Node.js environments, where extensive dependency trees create substantial optimization opportunities. Experimental results demonstrate that ServerlessPGO reduces cold start times by up to 71.7% for complex web frameworks in multi-concurrency scenarios. The framework achieves consistent improvements across various resource configurations, from 256MB to 2048MB allocations. ServerlessPGO operates through three mechanisms: reducing package sizes by up to 88.4%, consolidating file counts from thousands to under ten, and pre-computing module resolutions. The approach complements existing optimization techniques including container reuse and layer technology. Our evaluation reveals non-linear relationships between resource allocation and optimization effectiveness, with moderate configurations often outperforming maximum allocations. While current implementation targets Node.js on Alibaba Cloud Function Compute, the underlying principles apply across runtime environments. ServerlessPGO provides a practical solution for reducing cold start latency while maintaining compatibility with standard serverless deployment workflows.</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

ServerlessPGO: enhancing serverless cold starts through PGO

  • 期刊论文

  • 发表日期：

    2025年9月29日

查看全文
  • Yu Liu,
  • Li Fu,
  • Ma Penghui

摘要

The cold start latency issue in serverless computing remains a significant challenge, especially for large or dependency-heavy functions. This study introduces ServerlessPGO, which applies Profile-Guided Optimization (PGO) techniques to serverless architectures. ServerlessPGO analyzes runtime execution patterns to optimize dependencies and code paths, reducing function initialization overhead. Our implementation focuses on Node.js environments, where extensive dependency trees create substantial optimization opportunities. Experimental results demonstrate that ServerlessPGO reduces cold start times by up to 71.7% for complex web frameworks in multi-concurrency scenarios. The framework achieves consistent improvements across various resource configurations, from 256MB to 2048MB allocations. ServerlessPGO operates through three mechanisms: reducing package sizes by up to 88.4%, consolidating file counts from thousands to under ten, and pre-computing module resolutions. The approach complements existing optimization techniques including container reuse and layer technology. Our evaluation reveals non-linear relationships between resource allocation and optimization effectiveness, with moderate configurations often outperforming maximum allocations. While current implementation targets Node.js on Alibaba Cloud Function Compute, the underlying principles apply across runtime environments. ServerlessPGO provides a practical solution for reducing cold start latency while maintaining compatibility with standard serverless deployment workflows.