Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transformer-Based Model for Cold Start Mitigation in FaaS Architecture

Alexandre Savi Fayam Mbala Mouen, Jerry Lacmou Zeutouo, Vianney Kengne Tchendji

TL;DR

This work tackles cold-start latency in Function-as-a-Service (FaaS) by forecasting invocation patterns with a Transformer-based time-series model and integrating this forecast into proactive resource management. It combines DBSCAN-based workload clustering, a Transformer encoder-decoder for time-series forecasting, and OpenWhisk integration to adapt prewarmed containers and idle windows. Experimental results on Azure Functions data show substantial reductions in cold-start delay and frequency, including up to $79\%$ improvement in startup latency over conventional methods, and improved predictive accuracy over LSTM baselines. The work highlights the practicality of transformer-based proactive scheduling for latency-sensitive serverless workloads, while acknowledging limitations related to data granularity and platform-specific constraints and outlining directions for future improvements such as more advanced architectures and reinforcement learning integration.

Abstract

Serverless architectures, particularly the Function as a Service (FaaS) model, have become a cornerstone of modern cloud computing due to their ability to simplify resource management and enhance application deployment agility. However, a significant challenge remains: the cold start problem. This phenomenon occurs when an idle FaaS function is invoked, requiring a full initialization process, which increases latency and degrades user experience. Existing solutions for cold start mitigation are limited in terms of invocation pattern generalization and implementation complexity. In this study, we propose an innovative approach leveraging Transformer models to mitigate the impact of cold starts in FaaS architectures. Our solution excels in accurately modeling function initialization delays and optimizing serverless system performance. Experimental evaluation using a public dataset provided by Azure demonstrates a significant reduction in cold start times, reaching up to 79\% compared to conventional methods.

Transformer-Based Model for Cold Start Mitigation in FaaS Architecture

TL;DR

This work tackles cold-start latency in Function-as-a-Service (FaaS) by forecasting invocation patterns with a Transformer-based time-series model and integrating this forecast into proactive resource management. It combines DBSCAN-based workload clustering, a Transformer encoder-decoder for time-series forecasting, and OpenWhisk integration to adapt prewarmed containers and idle windows. Experimental results on Azure Functions data show substantial reductions in cold-start delay and frequency, including up to improvement in startup latency over conventional methods, and improved predictive accuracy over LSTM baselines. The work highlights the practicality of transformer-based proactive scheduling for latency-sensitive serverless workloads, while acknowledging limitations related to data granularity and platform-specific constraints and outlining directions for future improvements such as more advanced architectures and reinforcement learning integration.

Abstract

Serverless architectures, particularly the Function as a Service (FaaS) model, have become a cornerstone of modern cloud computing due to their ability to simplify resource management and enhance application deployment agility. However, a significant challenge remains: the cold start problem. This phenomenon occurs when an idle FaaS function is invoked, requiring a full initialization process, which increases latency and degrades user experience. Existing solutions for cold start mitigation are limited in terms of invocation pattern generalization and implementation complexity. In this study, we propose an innovative approach leveraging Transformer models to mitigate the impact of cold starts in FaaS architectures. Our solution excels in accurately modeling function initialization delays and optimizing serverless system performance. Experimental evaluation using a public dataset provided by Azure demonstrates a significant reduction in cold start times, reaching up to 79\% compared to conventional methods.

Paper Structure

This paper contains 19 sections, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. FaaS & Cold Start
  3. Literature Review
  4. Techniques for Reducing Cold Start Delay
  5. Techniques for Reducing Cold Start Frequency
  6. Proposed Approach
  7. Evaluation
  8. Experimental Framework and Model Configuration
  9. Data preprocessing and clustering with DBSCAN (Figure 3)
  10. Transformer model setup and hyperparameters
  11. Proactive Cold Start management with OpenWhisk (Figure 4)
  12. Reducing Cold Start Delay
  13. Reducing Cold Start Frequency
  14. General Discussion of Results and Implications for FaaS Environments
  15. Model limitations and edge cases
  16. ...and 4 more sections

Figures (3)

  • Figure 1: FaaS execution flow vahidinia_mitigating_2023
  • Figure 3: Global Data Processing Pipeline for Training the Transformer Model
  • Figure 4: Proposed Approach: Transformer-Based Model to Mitigate Cold Start