Adaptive Mechanism for Collaborative Service Deployment and Task Scheduling in Smart Cities

0
0
0

Adaptive Mechanism for Collaborative Service Deployment and Task Scheduling in Smart Cities

Introduction

The rapid development of mobile communication technologies, particularly 5G, has led to the emergence of new applications such as autonomous driving, virtual reality, augmented reality (AR), digital twins, and the metaverse. These applications demand high computational power and low latency, which traditional cloud computing struggles to provide due to bandwidth limitations and high latency. Multi-access edge computing (MEC) addresses these challenges by decentralizing computing and storage resources, placing them closer to end-user devices at the network edge. This approach reduces communication delays and enhances service responsiveness.

However, deploying MEC in smart cities introduces several challenges:

  1. Resource Constraints: Edge servers have limited computational and storage capacities.
  2. Diverse Service Requests: Different applications require specialized services, increasing complexity.
  3. Server Overload: High demand can lead to unbalanced workloads across edge servers.

To tackle these issues, this paper proposes an adaptive mechanism for collaborative service deployment and task scheduling in smart city environments. The goal is to optimize system latency and load balancing while considering task priorities.

System Model and Problem Definition

Network Architecture

The system consists of three layers:

  1. Cloud Data Center Layer: Provides centralized computing resources.
  2. Edge Layer: Comprises distributed edge servers that handle localized processing.
  3. User Device Layer: Includes mobile devices and sensors generating service requests.

Edge servers form a network where they can communicate directly, sharing bandwidth and computational resources. Each server has finite storage and processing capabilities, requiring efficient resource allocation.

Service Deployment Model

Services must be deployed on edge servers to process incoming tasks. Key constraints include:
• Minimum Deployment: Each service must be deployed at least once.

• Storage Limits: The total storage used by services on a server cannot exceed its capacity.

A resource pool is defined for each task, consisting of servers that host the required service.

Task Processing Model

Tasks are characterized by:
• Service Type: Determines which edge servers can process the task.

• Data Size: Affects transmission time between devices and servers.

• Computational Load: Influences processing time on edge servers.

• Priority: Higher-priority tasks should be processed faster.

The total task delay consists of:

  1. Transmission Delay: Time taken to send data from the device to the server.
  2. Processing Delay: Time required for the server to execute the task.
  3. Queueing Delay: Time spent waiting in the server’s task queue.

To ensure fairness, a weighted average response time (AWRT) is used, where higher-priority tasks contribute more to the overall delay metric.

Service-Level Agreement (SLA)

A priority-aware SLA is introduced to penalize delays differently based on task importance:
• Ideal Delay Threshold: Below this, no penalty is applied.

• Maximum Tolerable Delay: Beyond this, the penalty is maximized.

• Linear Penalty: Applied between the ideal and maximum thresholds.

This approach ensures that critical tasks (e.g., emergency services) receive stricter delay constraints than less critical ones (e.g., AR applications).

Load Balancing

A Gini coefficient measures server load inequality, where lower values indicate better balance. The goal is to minimize this coefficient while optimizing system latency.

Optimization Problem

The joint optimization of service deployment and task scheduling is formulated as a multi-objective problem, combining:

  1. System Latency: Minimizing AWRT.
  2. Load Balancing: Reducing the Gini coefficient.

The final objective function is a weighted sum of these two metrics.

Proposed Solution: HA_IBiEO Algorithm

The HA_IBiEO algorithm operates in a hierarchical time framework, separating service deployment (long-term decisions) and task scheduling (short-term decisions).

Key Components

  1. Service Deployment (HA Algorithm)
    • Global Access Rate: Frequently accessed services are deployed on all servers.

    • Local Access Rate: Less frequent services are selectively deployed based on historical demand.

    • Priority Consideration: Higher-priority services are given preference in deployment.

  2. Task Scheduling (IBiEO Algorithm)
    • Binary Equilibrium Optimizer (BiEO): A metaheuristic algorithm inspired by dynamic equilibrium models.

    • Improved Transfer Function: Combines V-shaped and U-shaped functions to enhance exploration and exploitation.

    • Priority-Based Queueing: Higher-priority tasks are processed first, reducing their waiting time.

Algorithm Workflow

  1. Initialization: Randomly deploy services and initialize task scheduling parameters.

  2. Service Deployment Update (Every τ Time Slots):
    • Identify frequently accessed services.

    • Adjust deployments based on server storage limits.

  3. Task Scheduling (Every Time Slot):
    • Use IBiEO to assign tasks to servers, considering:

    ◦ Current server load.

    ◦ Task priority.

    ◦ Transmission and processing delays.

  4. Dynamic Adjustment:
    • If a task cannot be processed due to resource constraints, it is reassigned to another eligible server.

    • Queueing order is adjusted dynamically to prioritize urgent tasks.

Advantages of HA_IBiEO

• Adaptive Service Deployment: Adjusts based on real-time demand patterns.

• Efficient Task Scheduling: Minimizes delays while balancing server loads.

• Priority Awareness: Ensures critical tasks meet their SLAs.

Experimental Evaluation

Dataset and Setup

The algorithm was tested using:
• Shanghai Telecom Dataset: Contains real-world base station locations and user access logs.

• Simulated Task Requests: Generated based on service popularity and priority distributions.

Key parameters:
• Edge Servers: 10, with varying computational resources.

• Task Priorities: Uniformly distributed from 1 (lowest) to 5 (highest).

• Maximum Tolerable Delay: Adjusted based on task priority.

Performance Metrics

  1. System Overhead: Combines latency and load balancing.
  2. AWRT: Measures the average delay weighted by task priority.
  3. Load Balancing (Gini Coefficient): Evaluates server workload distribution.
  4. Timeout Rate: Percentage of tasks exceeding their maximum tolerable delay.

Comparative Analysis

HA_IBiEO was compared against four baseline algorithms:

  1. LFU_IBiEO: Least Frequently Used service deployment + IBiEO scheduling.
  2. HA_BiEO: Proposed service deployment + standard BiEO scheduling.
  3. Greedy: Shortest-distance-based task scheduling.
  4. Pheu_SA: Access-based service deployment + simulated annealing scheduling.

Key Findings

  1. System Overhead
    • HA_IBiEO consistently achieved the lowest overhead, outperforming LFU_IBiEO by 6.94%, HA_BiEO by 17.03%, Pheu_SA by 20.78%, and Greedy by 25.96%.

  2. Load Balancing
    • HA_IBiEO maintained the most balanced server loads, with a Gini coefficient 17.02% lower than LFU_IBiEO and 39.19% lower than Greedy.

  3. AWRT
    • HA_IBiEO reduced AWRT by 20.05% compared to LFU_IBiEO and 69.43% compared to Greedy.

  4. Timeout Rate
    • HA_IBiEO was the only algorithm keeping timeout rates below 50% under high task arrival rates.

Impact of Key Parameters

  1. Maximum Tolerable Delay
    • As delay thresholds increased, system overhead decreased, but AWRT slightly increased due to relaxed constraints.

  2. Edge Server Computational Resources
    • Increasing resources improved performance, but gains diminished beyond a certain point (20 GHz in this study).

  3. Task Arrival Rate
    • Under high arrival rates, HA_IBiEO maintained stable performance, while other algorithms struggled with timeout rates exceeding 50%.

Conclusion

This paper presents an adaptive mechanism for joint service deployment and task scheduling in smart city MEC environments. The HA_IBiEO algorithm dynamically adjusts service placements and task assignments to minimize system latency and load imbalance, while respecting task priorities.

Key contributions include:

  1. Priority-Aware SLA: Ensures critical tasks meet strict delay requirements.
  2. Hierarchical Optimization: Separates long-term service deployment and short-term task scheduling.
  3. Improved BiEO Algorithm: Enhances exploration and exploitation in task scheduling.

Experimental results demonstrate that HA_IBiEO outperforms existing methods, reducing AWRT by 12.35% and improving load balancing by 14.47%.

Future Work

  1. Spatiotemporal Prediction: Incorporate user mobility patterns to anticipate demand shifts.
  2. Additional QoS Metrics: Extend the model to include energy consumption and operational costs.
  3. Scalability Testing: Evaluate performance in larger, more heterogeneous edge networks.

This work provides a robust foundation for optimizing MEC in smart cities, ensuring efficient resource utilization and high-quality service delivery.

doi.org/10.19734/j.issn.1001-3695.2024.05.0143

Was this helpful?

0 / 0