[Paper Review] Distance-Guided GA-Based Approach to Distributed Data-Intensive Web Service Composition

Soheila Sadeghiram, Hui Ma+1 more|arXiv (Cornell University)|Jan 16, 2019

Service-Oriented Architecture and Web Services23 references4 citations

TL;DR

This paper proposes a novel Genetic Algorithm (GA)-based approach for distributed data-intensive Web service composition (DWSC), integrating domain-specific heuristics—Longest Common Subsequence (LCS) and service distance—into crossover operators and a local search technique. The method significantly improves solution quality by preserving functional building blocks and optimizing for QoS, with distance-guided LCS crossover outperforming baseline and index-based methods across benchmark datasets.

ABSTRACT

Distributed computing which uses Web services as fundamental elements, enables high-speed development of software applications through composing many interoperating, distributed, re-usable, and autonomous services. As a fundamental challenge for service developers, service composition must fulfil functional requirements and optimise Quality of Service (QoS) attributes, simultaneously. On the other hand, huge amounts of data have been created by advances in technologies, which may be exchanged between services. Data-intensive Web services are of great interest to implement data-intensive processes. However, current approaches to Web service composition have omitted either the effect of data, or the distribution of services. Evolutionary Computing (EC) techniques allow for the creation of compositions that meet all the above factors. In this paper, we will develop Genetic Algorithm (GA)-based approach for solving the problem of distributed data-intensive Web service composition (DWSC). In particular, we will introduce two new heuristics, i.e. Longest Common Subsequence(LCS) distance of services, in designing crossover operators. Additionally, a new local search technique incorporating distance of services will be proposed.

Motivation & Objective

• To address the NP-hard challenge of distributed data-intensive Web service composition (DWSC) by integrating functional correctness and Quality of Service (QoS) optimization.
• To overcome limitations of existing Web service composition methods that neglect data volume and service distribution effects.
• To enhance genetic algorithm performance in DWSC by incorporating domain-specific heuristics into crossover and local search operators.
• To develop a fully automated, scalable solution for large-scale, data-intensive service compositions in distributed environments.
• To empirically validate the superiority of heuristic-guided crossover operators over general-purpose or index-based alternatives.

Proposed method

• Designs three problem-specific crossover operators using LCS and service distance heuristics to preserve functional building blocks and improve solution diversity.
• Introduces a distance-guided two-point crossover that balances exploration and exploitation by considering geographical and communication costs between services.
• Proposes a local search technique incorporating service distance to refine solutions and enhance convergence to high-quality optima.
• Employs a fitness function that combines QoS attributes (e.g., response time, cost) with data transfer size and communication overhead between distributed services.
• Uses benchmark datasets (WSC08 and WSC09) to evaluate performance across multiple runs, comparing against a baseline index-based crossover from prior WSC work.
• Applies statistical significance testing (pairwise comparisons) to validate performance differences across methods.

Experimental results

Research questions

• RQ1 How do LCS and service distance heuristics improve the quality of solutions in distributed data-intensive Web service composition?
• RQ2 Can distance-guided crossover operators outperform general-purpose or index-based crossover operators in DWSC?
• RQ3 What is the impact of incorporating domain-specific knowledge into GA crossover and local search on solution convergence and QoS optimization?
• RQ4 How does the proposed local search technique enhance solution quality in comparison to standard GA approaches?
• RQ5 Which crossover variant—GA-I, GA-II, or GA-LCS—produces the most consistent and optimal results across diverse DWSC datasets?

Key findings

• The distance-guided LCS crossover (GA-LCS) achieved the best overall performance, producing significantly better fitness values than all other methods across all benchmark datasets.
• GA-LCS consistently outperformed the baseline index-based crossover from [20], with lower fitness values indicating superior QoS optimization.
• For dataset WSC08-2, GA-LCS achieved a fitness of 0.46 ± 0.04, significantly better than the index crossover’s 0.51 ± 0.09.
• In WSC08-4, GA-LCS achieved 0.41 ± 0.01, the lowest among all methods, demonstrating strong convergence and solution quality.
• GA-II, which increases offspring diversity by increasing distance from parents, outperformed GA-I, indicating that higher diversity improves solution quality.
• Statistical pairwise comparisons confirmed that distance-guided crossovers consistently produced significantly better results than the index-based method, validating the effectiveness of domain-specific heuristics.