Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Distributed Edge Computing Task Allocation with Network Effects

Henry Abrahamson, Yongho Kim, Seongha Park, Ermin Wei

TL;DR

This work addresses task allocation across distributed edge computing nodes under heterogeneous hardware and time-varying QoS requirements. It formulates the problem as a distributed optimization and solves it with a dual-descent method, enabling in-network coordination without a central controller. Key contributions include a novel distributed task-allocation formulation with per-task QoS constraints, demonstration that standard distributed optimization can solve the problem in a decentralized fashion, and Sage-data–driven evaluation showing fast adaptation to dynamic conditions. The findings highlight practical impact for reactive, multi-tenant edge networks performing urgent sensing tasks with heterogeneous devices.

Abstract

Field-deployable edge computing nodes form a network and are used to complete scientific tasks for remote sensing and monitoring. The networked nodes collectively decide which scientific applications to run while they are constrained by various factors, such as differing hardware constraints from heterogeneous nodes and time-varying quality of service (QoS) requirements. We model the problem of task allocation as an optimization problem that maximizes the QoS, subject to the constraints. We solve the optimization problem using a dual-descent method, which can be easily implemented in a distributed way subject to the communication constraints of the network. Using a simulation that uses real-world data collected from Sage, a distributed sensor network, we analyze our policy's performance in dynamic situations where the required QoS and the nodes' capabilities change, and verify that it can adapt and return a feasible solution while accounting for those changes.

Distributed Edge Computing Task Allocation with Network Effects

TL;DR

This work addresses task allocation across distributed edge computing nodes under heterogeneous hardware and time-varying QoS requirements. It formulates the problem as a distributed optimization and solves it with a dual-descent method, enabling in-network coordination without a central controller. Key contributions include a novel distributed task-allocation formulation with per-task QoS constraints, demonstration that standard distributed optimization can solve the problem in a decentralized fashion, and Sage-data–driven evaluation showing fast adaptation to dynamic conditions. The findings highlight practical impact for reactive, multi-tenant edge networks performing urgent sensing tasks with heterogeneous devices.

Abstract

Field-deployable edge computing nodes form a network and are used to complete scientific tasks for remote sensing and monitoring. The networked nodes collectively decide which scientific applications to run while they are constrained by various factors, such as differing hardware constraints from heterogeneous nodes and time-varying quality of service (QoS) requirements. We model the problem of task allocation as an optimization problem that maximizes the QoS, subject to the constraints. We solve the optimization problem using a dual-descent method, which can be easily implemented in a distributed way subject to the communication constraints of the network. Using a simulation that uses real-world data collected from Sage, a distributed sensor network, we analyze our policy's performance in dynamic situations where the required QoS and the nodes' capabilities change, and verify that it can adapt and return a feasible solution while accounting for those changes.
Paper Structure (8 sections, 7 equations, 6 figures, 2 tables)

This paper contains 8 sections, 7 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Formulation
  3. Motivating Example: Wildfire Detection and Monitoring
  4. Problem Formulation
  5. Numerical Simulation
  6. Data Preparation with Sage
  7. Simulation Structure
  8. Simulation Runs and Results

Figures (6)

  • Figure 1: Wildfire and smoke detection as a use case: geographically grouped nodes run the smokey AI model dewangan2022figlib on visual and infrared camera images for early wildfire detection.
  • Figure 2: An example network. Each different colored dotted circle represents a different node group, with different node groups for task 1 (left) versus task 2 (right). Nodes within a group can communicate freely with each other.
  • Figure 3: The network used in our simulation. The wildfire monitoring task (task 1) uses the node groups on the left, while the cloud analysis task (task 2) uses the node groups on the right. Nodes 3 and 5, indicated in yellow, are higher-performance Nvidia Jetson devices, while nodes 1, 2, 4, and 6, indicated in white, are lower-end Raspberry Pi devices.
  • Figure 4: The number of tasks ran on each node group for the degradation simulation. The left shows the node groups for task 1, while the right shows the node groups for task 2. The dynamic and static policy overlay each other at first, but the static policy fails to uphold the QoS minimum in high degradation while the dynamic policy maintains it.
  • Figure 5: The number of tasks ran on each node for the degradation simulation. As nodes 3 and 6 start falling below the minimum QoS requirements, other nodes in their node groups reallocate resources to fill the gaps.
  • ...and 1 more figures