Recent Advances in Collaborative Scheduling of Computing Tasks in an Edge Computing Paradigm

Shichao Chen, Qijie Li, Mengchu Zhou, Abdullah Abusorrah2021 · DOI

TL;DR

This paper reviews recent advancements in collaborative task scheduling for edge computing, addressing the challenge of efficiently processing data from the Internet of Things (IoT). It proposes a framework for optimizing task offloading and scheduling across edge devices, edge servers, and cloud centers to minimize latency and energy consumption, crucial for real-time IoT applications.

What Problem Does This Paper Address?

The explosion of Internet of Things (IoT) devices generates massive amounts of data. While cloud computing offers powerful processing capabilities, it suffers from high transmission delays and network bandwidth strain. Edge computing aims to alleviate these issues by processing data closer to the source. However, simply offloading all tasks to edge servers can lead to their overload, causing delays and wasted resources on idle edge devices. This paper tackles the problem of collaborative scheduling – how to intelligently distribute computing tasks among edge devices, edge servers, and cloud centers to achieve optimal performance (low latency, high efficiency) and resource utilization.

Methods

The paper provides a comprehensive survey of existing research, categorizing edge computing scenarios and task types. It then formulates an optimization problem for computation offloading, considering multiple objectives like minimizing delay and energy consumption. The core of the paper lies in its review and classification of various collaborative scheduling methods. These methods range from traditional algorithms like Min-Min and Max-Min to more advanced intelligent heuristic approaches such as Genetic Algorithms (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization. The paper also discusses models for analyzing task execution, communication, and offloading, including queueing models (M/M/1, M/M/n, M/M/∞) to represent system dynamics.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • The "Edge" is Not Monolithic: Edge computing isn't just about edge servers; it's a spectrum involving edge devices, edge servers, and the cloud, each with different resource capabilities.
  • Collaborative Scheduling is Key: Simply offloading to the nearest edge server isn't always optimal. Intelligent scheduling across the entire edge-cloud continuum is necessary.
  • Task Characteristics Matter: The nature of a computing task (e.g., its size, divisibility, delay sensitivity) dictates the best offloading strategy.
  • Multi-Objective Optimization is Crucial: Balancing latency and energy consumption is a fundamental challenge, as these objectives often conflict.
  • Heuristics Dominate: Due to the NP-hard nature of optimal scheduling, heuristic and intelligent optimization algorithms are widely used to find near-optimal solutions efficiently.

Key Results

The paper doesn't present new experimental results but rather synthesizes existing research. It highlights that various scheduling algorithms have been proposed to achieve:

  • Minimal Delay Time: Algorithms like greedy approaches and Lyapunov function-based methods aim to reduce task completion time. For instance, some studies show average response time reductions of 20-30% compared to simpler methods.
  • Minimal Energy Consumption: Techniques involving particle swarm optimization and dynamic voltage frequency scaling are used to reduce energy usage, with some showing stable convergence and optimal adaptability.
  • Balancing Delay and Energy: Algorithms that consider both objectives often involve trade-offs, with some achieving reductions of around 30% in overall cost (combining delay and energy).

Limitations and Future Work

The authors identify several critical areas for future research:

  • Handling Emergencies: The current scheduling models often don't account for device failures or network outages, which can significantly impact task execution. Future work should incorporate failure probabilities into scheduling decisions.
  • Expanding Optimization Objectives: Beyond delay and energy, other Quality of Service (QoS) indicators like security, privacy, and user experience need to be integrated into scheduling goals.
  • Data Security and Delay Trade-offs: Developing lightweight security protocols and encryption algorithms that don't introduce significant delays or energy overhead is a challenge.
  • Any-Time Task Scheduling: The need for fast scheduling algorithms that can produce a feasible solution quickly, with the option for further optimization if more time is available, is crucial for real-time applications.
  • Incorporating More Factors: Factors like offloading ratio and the specific characteristics of Deep Learning (DL) models (especially for partial offloading) need to be better integrated into optimization goals.
  • Balancing DL Model Offloading: Optimizing the partial offloading of DL models for training and inference, considering resource constraints on edge devices, is a complex but important future direction.

Summary updated

BibTeX
@misc{chen2021recent,
  author = {Shichao Chen and Qijie Li and Mengchu Zhou and Abdullah Abusorrah},
  title = {{Recent Advances in Collaborative Scheduling of Computing Tasks in an Edge Computing Paradigm}},
  year = {2021},
  doi = {10.3390/s21030779},
}
Figure