A Comparative Evaluation of Population-based Optimization Algorithms for Workflow Scheduling in Cloud-Fog Environments (2012.00176v2)

Abstract: This work presents a comparative evaluation of four population-based optimization algorithms for workflow scheduling in cloud-fog environments. These algorithms are as follows: Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Differential Evolution (DE) and GA-PSO. This work also provides the motivational groundwork for the weighted sum objective function for the workflow scheduling problem and develops this function based on three objectives: makespan, cost and energy. The recently proposed FogWorkflowSim is used as the simulation environment with the aforementioned objectives serving performance metrics. Results show that hybrid combination of the GA-PSO algorithm exhibits slightly better than the standard algorithms. Future work will include expansion of the workflows used by increasing the number of tasks as well as adding some more workflows. The addition of some more objectives to the weighted objective function will also be pursued

Comparative Analysis of Workflow Scheduling Optimization in Cloud-Fog Environments

The paper "A Comparative Evaluation of Population-based Optimization Algorithms for Workflow Scheduling in Cloud-Fog Environments" explores the performance of various population-based optimization algorithms applied to workflow scheduling within cloud-fog computational infrastructures. Specifically, it investigates Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Differential Evolution (DE), and a hybrid GA-PSO algorithm using the Fog WorkflowSim simulation environment which captures the dynamics of a cloud-fog processing architecture.

Core Contributions

The primary contributions manifested in this paper are manifold:

1. Weighted Sum Objective Function: The authors introduce a composite objective function encapsulating makespan, cost, and energy consumption as critical metrics for evaluating workflow scheduling. This tri-metric framework is tailored towards optimizing these conflicting objectives synergistically, particularly recognizing the energy limitations posed by fog servers compared to cloud servers.
2. Algorithmic Diversification: Differential Evolution (DE), previously unexplored in workflow scheduling in cloud-fog contexts, is investigated. Furthermore, the hybrid GA-PSO approach, reputed for its synergistic advantages in cloud environments, is adapted to cloud-fog settings.
3. Implementation in FogWorkflowSim: The paper integrates DE and GA-PSO into the FogWorkflowSim toolkit, allowing for nuanced comparative analyses of the algorithms using realistic workload models.

Methodology

The paper methodically employs a discrete encoding strategy for task-resource assignments, evaluating performance based on makespan, cost, and energy consumption. The optimization algorithms are systematically applied to predefined workflow models (Montage, Epigenomics, CyberShake), which are instantiated as Directed Acyclic Graphs (DAGs), representing complex computational task dependencies. Each algorithm's efficacy is quantified and compared under controlled simulation environments with specific parameter settings critical for replicability and consistency in experimentation.

Findings

The empirical investigations revealed distinct performance patterns across workflows and algorithmic configurations:

• GA-PSO Performance: Across various workflows, the GA-PSO hybrid frequently demonstrated superior performance, particularly in terms of maintaining favorable balances across the performance metrics. This suggests the hybrid approach effectively mitigates some of the limitations seen in standalone PSO and GA, such as PSO's premature convergence.
• Algorithm Comparisons: While GA-PSO generally excelled, the DE algorithm presented unique advantages, particularly in reducing costs by optimizing task allocation strategies, albeit at the expense of higher energy consumption, especially as task scales increased.
• PSO Limitations: PSO generally exhibited suboptimal performance, reflecting its documented susceptibility to local minima in optimization landscapes without diverse initialization strategies.

Theoretical and Practical Implications

The paper enriches the theoretical discourse on multi-objective optimization in distributed environments by providing empirical evidence favoring hybrid algorithm configurations. Practically, the insights facilitate developments in smart resource allocation mechanisms, potentially informing adaptive scheduling frameworks in future iterations of fog computing infrastructures.

Future Directions

Future research avenues, as indicated by the authors, include extending workflow tasks to broader horizons, potentially up to 2000 tasks, and exploring additional workflow types such as LIGO and SIPHT. Incorporating reliability metrics, fault tolerance, and adhering to budgetary and deadline constraints in future weighted sum objectives is also projected, providing a more comprehensive solution vector for dynamic cloud-fog resource management.

In summary, the paper constitutes a valuable contribution to the domain of cloud-fog workflow scheduling by offering a rigorous computational evaluation of multiple optimization methodologies, thereby laying the groundwork for more refined and adaptive task scheduling solutions in heterogeneous distributed environments.