Core Concepts
Developing a Multi-Objective Genetic Algorithm to optimize service placement in edge-to-cloud AR/VR systems, focusing on minimizing response time and maximizing system reliability.
Abstract
The content discusses the challenges of optimizing service placement in edge-to-cloud AR/VR systems. It introduces a Multi-Objective Genetic Algorithm (MOGA) to address these challenges by minimizing response time and maximizing system reliability. The article outlines the infrastructure model, service model, and optimization approach. It also presents experimental setups, evaluation scenarios, and the configuration setup of MOGA.
Infrastructure Model:
- Three-tier infrastructure: Access points, edge nodes, cloud.
- Characteristics of computing nodes: Computation capacity, memory capacity, disk capacity.
- Communication links among users, helpers, and computing nodes.
Service Model:
- Multiple AR/VR services with service components and versions.
- Directed Acyclic Graph representation for interdependent service components.
- Data transmission delay model based on bandwidth and data size.
Optimization Approach:
- Objective function to minimize response time and maximize hardware/software reliability.
- Chromosome encoding for mapping service components to computing nodes.
- Fitness function based on response time and reliability scores.
- Selection operator using tournament strategy.
- Crossover and mutation operators for genetic operations.
- Healing operator to ensure constraint satisfaction.
Experimental Setup:
- Tailor-made simulator implemented in Node.js for precise simulation of infrastructure and services.
- Evaluation scenarios: Small-scale, medium-scale, large-scale, xLarge-scale.
- Configuration setup of MOGA using grid-based tuning strategy.
Other Scheduling Algorithms:
- Task Continuation Affinity (TCA), Least Required CPU (LRC), Most Data Size (MDS), Most Reliability (MR), Most Powerful (MP), Least Powerful (LP).
Optimal Configurations:
For different scales:
- Population size: 200 for small-scale, 300 for medium-scale, 400 for large-scale, 500 for xLarge-scale.
- Crossover rate: Ranges from 60% to 80% based on problem size.
- Mutation rate: Fixed at 1% for all scenarios.
- Selection size: Ranges from 20 to 50 based on problem size.
- Number of iterations: Estimated using equations based on problem characteristics.
Stats
MOGA can reduce response time by an average of 67% compared to heuristic methods.