Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lumos: Performance Characterization of WebAssembly as a Serverless Runtime in the Edge-Cloud Continuum

Cynthia Marcelino, Noah Krennmair, Thomas Pusztai, Stefan Nastic

TL;DR

Lumos tackles the challenge of understanding WebAssembly as a serverless runtime in the Edge-Cloud Continuum by introducing a performance model and a modular benchmarking framework. It systematically characterizes workload-, system-, and environment-driven performance factors and compares Wasm (AoT and interpreted) against Docker containers across diverse edge workloads. Key findings show that AoT Wasm images are significantly smaller and can reduce cold-start latency, while interpreted Wasm suffers substantial warm-latency and I/O/serialization overhead, with containers often delivering better scalability and throughput under concurrency. The work provides actionable insights for deploying serverless workloads at the edge and outlines a path for extending Lumos to richer environments and real traces, enabling more informed trade-offs between portability, isolation, and performance.

Abstract

WebAssembly has emerged as a lightweight and portable runtime to execute serverless functions, particularly in heterogeneous and resource-constrained environments such as the Edge Cloud Continuum. However, the performance benefits versus trade-offs remain insufficiently understood. This paper presents Lumos, a performance model and benchmarking tool for characterizing serverless runtimes. Lumos identifies workload, system, and environment-level performance drivers in the Edge-Cloud Continuum. We benchmark state-of-the-art containers and the Wasm runtime in interpreted mode and with ahead-of-time compilation. Our performance characterization shows that AoT-compiled Wasm images are up to 30x smaller and decrease cold-start latency by up to 16% compared to containers, while interpreted Wasm suffers up to 55x higher warm latency and up to 10x I/O-serialization overhead.

Lumos: Performance Characterization of WebAssembly as a Serverless Runtime in the Edge-Cloud Continuum

TL;DR

Lumos tackles the challenge of understanding WebAssembly as a serverless runtime in the Edge-Cloud Continuum by introducing a performance model and a modular benchmarking framework. It systematically characterizes workload-, system-, and environment-driven performance factors and compares Wasm (AoT and interpreted) against Docker containers across diverse edge workloads. Key findings show that AoT Wasm images are significantly smaller and can reduce cold-start latency, while interpreted Wasm suffers substantial warm-latency and I/O/serialization overhead, with containers often delivering better scalability and throughput under concurrency. The work provides actionable insights for deploying serverless workloads at the edge and outlines a path for extending Lumos to richer environments and real traces, enabling more informed trade-offs between portability, isolation, and performance.

Abstract

WebAssembly has emerged as a lightweight and portable runtime to execute serverless functions, particularly in heterogeneous and resource-constrained environments such as the Edge Cloud Continuum. However, the performance benefits versus trade-offs remain insufficiently understood. This paper presents Lumos, a performance model and benchmarking tool for characterizing serverless runtimes. Lumos identifies workload, system, and environment-level performance drivers in the Edge-Cloud Continuum. We benchmark state-of-the-art containers and the Wasm runtime in interpreted mode and with ahead-of-time compilation. Our performance characterization shows that AoT-compiled Wasm images are up to 30x smaller and decrease cold-start latency by up to 16% compared to containers, while interpreted Wasm suffers up to 55x higher warm latency and up to 10x I/O-serialization overhead.

Paper Structure

This paper contains 53 sections, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Research Challenges
  3. RC-1: How do workload, system, and environmental factors influence the performance of serverless functions across the Edge–Cloud Continuum?
  4. RC-2: How do serverless runtimes impact function performance in the Edge-Cloud Continuum?
  5. RC-3: How to characterize serverless function performance under different workload characteristics and heterogeneous environments, such as the Edge-Cloud Continuum?
  6. Serverless Performance Model for the Edge-Cloud Continuum
  7. Workload Performance Factors
  8. Function Heterogeneity
  9. Workflow Dependencies
  10. Function State and BaaS
  11. Serialization Overhead
  12. System Performance Factors
  13. Performance Isolation
  14. Function Isolation
  15. Function Invocation
  16. ...and 38 more sections

Figures (8)

  • Figure 1: Serverless Performance Model for the Edge-Cloud Continuum, highlighting performance factors: workload, system, and environment.
  • Figure 2: Serverless function lifecycle highlighting cold, warm, and stateful functions
  • Figure 3: Lumos Architecture Overview
  • Figure 4: Image sizes of serverless functions
  • Figure 5: Execution time comparison across I/O- and CPU-intensive functions for varying input sizes and group sizes.
  • ...and 3 more figures