[Paper Review] Cold Start Latency in Serverless Computing: A Systematic Review, Taxonomy, and Future Directions

Muhammed Golec, Guneet Kaur Walia+4 more|arXiv (Cornell University)|Oct 12, 2023

IoT and Edge/Fog Computing4 citations

TL;DR

This paper presents a systematic review and taxonomy of cold start latency in serverless computing, analyzing 32 studies to classify solutions into caching, application-level optimization, and AI/ML-based approaches. It identifies key challenges and future research directions, offering a comprehensive framework for reducing cold start latency and frequency in serverless environments.

ABSTRACT

Recently, academics and the corporate sector have paid attention to serverless computing, which enables dynamic scalability and an economic model. In serverless computing, users only pay for the time they actually use resources, enabling zero scaling to optimise cost and resource utilisation. However, this approach also introduces the serverless cold start problem. Researchers have developed various solutions to address the cold start problem, yet it remains an unresolved research area. In this article, we propose a systematic literature review on clod start latency in serverless computing. Furthermore, we create a detailed taxonomy of approaches to cold start latency, which we use to investigate existing techniques for reducing the cold start time and frequency. We have classified the current studies on cold start latency into several categories such as caching and application-level optimisation-based solutions, as well as Artificial Intelligence (AI)/Machine Learning (ML)-based solutions. Moreover, we have analyzed the impact of cold start latency on quality of service, explored current cold start latency mitigation methods, datasets, and implementation platforms, and classified them into categories based on their common characteristics and features. Finally, we outline the open challenges and highlight the possible future directions.

Motivation & Objective

• To address the critical challenge of cold start latency in serverless computing, which hinders performance in time-sensitive applications.
• To identify and categorize existing solutions from academic and industrial research to understand their approaches and effectiveness.
• To provide a systematic taxonomy of cold start mitigation techniques, including AI/ML, caching, and function state management.
• To highlight research gaps and propose future directions, such as improved AI models, transfer learning, and open datasets.
• To support researchers and practitioners in advancing serverless computing by synthesizing current knowledge and identifying actionable research opportunities.

Proposed method

• Conducted a systematic literature review using PRISMA guidelines, analyzing 100+ papers from academic and industrial sources.
• Applied the snowball technique to expand the initial set, resulting in a curated dataset of 32 papers focused on cold start solutions.
• Classified solutions into five main categories: caching and application-level optimization, AI/ML-based prediction, function state persistence, function fusion, and runtime architecture improvements.
• Evaluated solution effectiveness based on cold start latency reduction, frequency reduction, and implementation feasibility.
• Used five research questions to guide analysis, focusing on QoS impact, latency factors, solution classification, platform applicability, and open science practices.
• Reviewed journal publication trends and code/dataset sharing practices to assess research transparency and reproducibility.

Experimental results

Research questions

• RQ1 What is the relationship between cold start latency and Quality of Service (QoS) in serverless applications?
• RQ2 What are the key factors that influence cold start latency in serverless computing environments?
• RQ3 How are existing solutions to cold start latency classified, and what are their common characteristics and differences?
• RQ4 On which platforms and runtime environments have cold start mitigation solutions been implemented, and what is their distribution?
• RQ5 What is the extent of code and dataset sharing in the current literature on cold start solutions?

Key findings

• AI/ML-based models for predicting function execution and cold start timing show promise but suffer from low performance and lack of generalization.
• Caching and application-level optimization techniques are widely adopted and effective in reducing cold start frequency and latency.
• Function state persistence and fusion-based approaches show potential but face scalability issues when handling diverse or parallel function inputs.
• Quantum computing and transfer learning are emerging as promising enablers for improving AI model performance in cold start prediction.
• A significant gap exists in open-source datasets and reproducible code, limiting model training and benchmarking in cold start research.
• Current isolation and runtime architectures are not optimized for cold start mitigation, indicating a need for new design principles and scheduling strategies.