Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy-Efficient Deployment of Stateful FaaS Vertical Applications on Edge Data Networks

Claudio Cicconetti, Raffaele Bruno, Andrea Passarella

TL;DR

The paper tackles energy-efficient deployment of stateful FaaS on edge data networks in a 5G/EDN setting, addressing the difficulty of maintaining state efficiently on decentralized edge resources. It compares stateless FaaS with external state against a stateful FaaS paradigm using lightweight WebAssembly runtimes, and develops an end-to-end energy model that couples processing and network costs over dynamic workloads, expressed through metrics such as active nodes and per-app traffic. A simple best-fit defragmentation heuristic is proposed to mitigate fragmentation and migration overhead, and extensive trace-based simulations show that stateful FaaS can often be more energy-efficient than stateless under realistic state sizes and network energy costs, with performance sensitive to the defragmentation period and workload characteristics. The work provides actionable guidance for energy-aware edge orchestration and contributes open-source tooling to support reproducibility and further research in edge-native serverless deployments.

Abstract

5G and beyond support the deployment of vertical applications, which is particularly appealing in combination with network slicing and edge computing to create a logically isolated environment for executing customer services. Even if serverless computing has gained significant interest as a cloud-native technology its adoption at the edge is lagging, especially because of the need to support stateful tasks, which are commonplace in, e.g., cognitive services, but not fully amenable to being deployed on limited and decentralized computing infrastructures. In this work, we study the emerging paradigm of stateful Function as a Service (FaaS) with lightweight task abstractions in WebAssembly. Specifically, we assess the implications of deploying inter-dependent tasks with an internal state on edge computing resources using a stateless vs. stateful approach and then derive a mathematical model to estimate the energy consumption of a workload with given characteristics, considering the power used for both processing and communication. The model is used in extensive simulations to determine the impact of key factors and assess the energy trade-offs of stateless vs. stateful.

Energy-Efficient Deployment of Stateful FaaS Vertical Applications on Edge Data Networks

TL;DR

The paper tackles energy-efficient deployment of stateful FaaS on edge data networks in a 5G/EDN setting, addressing the difficulty of maintaining state efficiently on decentralized edge resources. It compares stateless FaaS with external state against a stateful FaaS paradigm using lightweight WebAssembly runtimes, and develops an end-to-end energy model that couples processing and network costs over dynamic workloads, expressed through metrics such as active nodes and per-app traffic. A simple best-fit defragmentation heuristic is proposed to mitigate fragmentation and migration overhead, and extensive trace-based simulations show that stateful FaaS can often be more energy-efficient than stateless under realistic state sizes and network energy costs, with performance sensitive to the defragmentation period and workload characteristics. The work provides actionable guidance for energy-aware edge orchestration and contributes open-source tooling to support reproducibility and further research in edge-native serverless deployments.

Abstract

5G and beyond support the deployment of vertical applications, which is particularly appealing in combination with network slicing and edge computing to create a logically isolated environment for executing customer services. Even if serverless computing has gained significant interest as a cloud-native technology its adoption at the edge is lagging, especially because of the need to support stateful tasks, which are commonplace in, e.g., cognitive services, but not fully amenable to being deployed on limited and decentralized computing infrastructures. In this work, we study the emerging paradigm of stateful Function as a Service (FaaS) with lightweight task abstractions in WebAssembly. Specifically, we assess the implications of deploying inter-dependent tasks with an internal state on edge computing resources using a stateless vs. stateful approach and then derive a mathematical model to estimate the energy consumption of a workload with given characteristics, considering the power used for both processing and communication. The model is used in extensive simulations to determine the impact of key factors and assess the energy trade-offs of stateless vs. stateful.
Paper Structure (11 sections, 5 equations, 13 figures, 1 table)

This paper contains 11 sections, 5 equations, 13 figures, 1 table.

Table of Contents

  1. Introduction
  2. Stateless vs. Stateful FaaS: Background and Motivation
  3. Stateful FaaS benchmarking
  4. System Model
  5. Stateless FaaS
  6. Stateful FaaS
  7. Performance Evaluation
  8. Related Work
  9. Edge computing support in 5G systems
  10. Application deployment
  11. Conclusions

Figures (13)

  • Figure 1: Simplified beyond 5G architecture of an EDN mobile users with EAS services.
  • Figure 2: Example of how to realize stateful processing with stateless FaaS.
  • Figure 3: Example of stateful FaaS.
  • Figure 4: Migration of a stateful FaaS runner from node $A$ to node $B$.
  • Figure 5: Deployment of a three-function chain (top) on two processing nodes through stateless FaaS (middle) and stateful FaaS (bottom).
  • ...and 8 more figures