Scholarlink

Function-as-a-Service (FaaS) has emerged one of the most innovative and affordable cloud service model for implementing serverless computing. FaaS has received much attention because of its scalability, resource efficiency, cost effectiveness, and ease of deployment features. However, the “scale to zero” feature of FaaS results in a problem called cold start latency, that is, a delay which occurs when the functions are invoked for the first time or after a period of inactivity. In this paper, we introduced an Uncertainty-Aware-based Random Forest (UARF) predictive pre-warming model which reduces the cold start latency. The UARF model uses the historical invocation patterns to generate the prediction probability and variance, which helps in taking decision whether to pre-warm or not. Based on these prediction, the UARF model, keeps the right number of containers pre-warmed. The experiment results on the OpenFaaS platform show that the proposed approach reduces the number of cold starts and cold start latency by 20% as compared to baseline OpenFaaS behavior.

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 Cold Start Latency in Serverless Computing Using Uncertainty-Aware-Based Random Forest Predictive Pre-warming Technique

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Manjeet Singh,
  • Man Mohan Singh Rauthan,
  • Sarita Negi

摘要

Function-as-a-Service (FaaS) has emerged one of the most innovative and affordable cloud service model for implementing serverless computing. FaaS has received much attention because of its scalability, resource efficiency, cost effectiveness, and ease of deployment features. However, the “scale to zero” feature of FaaS results in a problem called cold start latency, that is, a delay which occurs when the functions are invoked for the first time or after a period of inactivity. In this paper, we introduced an Uncertainty-Aware-based Random Forest (UARF) predictive pre-warming model which reduces the cold start latency. The UARF model uses the historical invocation patterns to generate the prediction probability and variance, which helps in taking decision whether to pre-warm or not. Based on these prediction, the UARF model, keeps the right number of containers pre-warmed. The experiment results on the OpenFaaS platform show that the proposed approach reduces the number of cold starts and cold start latency by 20% as compared to baseline OpenFaaS behavior.