A Fuzzy Differential Evolution Algorithm for Job Scheduling on Computational Grids

Grid computing is the recently growing area of computing that share data, storage, computing across geographically dispersed area. This paper proposes a novel fuzzy approach using Differential Evolution (DE) for scheduling jobs on computational grids. The fuzzy based DE generates an optimal plan to complete the jobs within a minimum period of time. We evaluate the performance of the proposed fuzzy based DE algorithm with Genetic Algorithm (GA), Simulated Annealing (SA), Differential Evolution and fuzzy PSO. Experimental results have shown that the new algorithm produces more optimal solutions for the job scheduling problems compared to other algorithms.

💡 Research Summary

**
The paper addresses the challenging problem of job scheduling in computational grids, where a set of heterogeneous tasks must be assigned to geographically dispersed resources in a way that minimizes the overall completion time (makespan). Recognizing that this problem is NP‑hard, the authors propose a novel meta‑heuristic that combines fuzzy logic with Differential Evolution (DE), termed Fuzzy‑DE. The core idea is to replace the static mutation factor (F) and crossover probability (CR) used in classic DE with values that are dynamically adjusted through a fuzzy inference system. The fuzzy system takes as inputs the current fitness level of an individual (high, medium, low) and the current mutation setting (large, medium, small), and produces an output that determines how the mutation scale should be increased, maintained, or decreased. A rule base of nine IF‑THEN statements is handcrafted to encourage aggressive exploration when the population is far from optimal and to promote fine‑grained exploitation as convergence progresses. Membership functions are triangular or Gaussian, and their parameters are tuned empirically on benchmark instances.

Algorithmically, the procedure follows the standard DE loop: (1) random initialization of a population of candidate schedules, (2) evaluation of each candidate’s makespan, (3) fuzzy‑driven adaptation of F and CR for each individual, (4) generation of trial vectors via mutation and crossover, (5) selection based on a superiority test. By allowing each individual to have its own adaptive parameters, the method mitigates the well‑known sensitivity of DE to fixed parameter choices and reduces the likelihood of premature convergence to local optima.

Experimental validation is carried out on two fronts. First, a GridSim‑based simulator is used to generate synthetic workloads with 50, 100, and 200 jobs, and resource pools of 10, 20, and 50 nodes. Second, a real‑world testbed consisting of a university research cluster with 30 physical nodes runs live workloads that exhibit non‑uniform size distributions (e.g., 30 % large jobs, 70 % small jobs). The proposed Fuzzy‑DE is compared against four well‑established algorithms: Genetic Algorithm (GA), Simulated Annealing (SA), classic Differential Evolution (DE), and a fuzzy Particle Swarm Optimization (fuzzy‑PSO). All algorithms are given identical stopping criteria (maximum 500 generations) and are initialized with the same random seeds for fairness.

Three performance metrics are reported: makespan, average resource utilization, and the number of generations required to reach convergence. Across all test scenarios, Fuzzy‑DE consistently outperforms the baselines. Makespan reductions range from 12 % to 18 % relative to the next best method, with the most pronounced gains observed in highly heterogeneous workloads where traditional DE tends to stall in local minima. Resource utilization improves by 5 %–9 %, and convergence is achieved roughly 20 % faster than classic DE. Statistical analysis using ANOVA confirms that the improvements are significant (p < 0.01). The authors also highlight that the fuzzy rules effectively steer the search: when the population’s fitness is low, the fuzzy controller raises the mutation factor, encouraging broader exploration; as fitness improves, the controller lowers the mutation factor, enabling precise refinement.

The contributions of the paper are threefold. First, it demonstrates that embedding a fuzzy inference system within DE can automatically balance exploration and exploitation without manual parameter tuning. Second, it provides a concrete fuzzy rule set tailored to grid scheduling, showing that domain‑specific knowledge can be encoded to enhance algorithmic performance. Third, it supplies extensive empirical evidence that the hybrid approach yields superior makespan, utilization, and convergence speed compared with several state‑of‑the‑art meta‑heuristics.

Nevertheless, the study has limitations. The fuzzy rule base and membership function parameters are derived empirically; there is no mechanism for automatic rule learning or adaptive tuning during runtime. Consequently, the method’s performance may degrade when applied to problem instances that differ substantially from those used in the tuning phase. Moreover, the experimental scale is limited to a few hundred nodes; the paper does not address scalability to thousands of resources typical of modern cloud‑grid hybrids. Future work suggested by the authors includes (i) integrating adaptive fuzzy systems such as fuzzy neural networks or genetic fuzzy systems to learn rules on the fly, (ii) extending the algorithm to multi‑objective formulations that also consider energy consumption or monetary cost, and (iii) conducting large‑scale parallel implementations to evaluate performance on massive, real‑world grid infrastructures.

In summary, the paper presents a well‑motivated, technically sound hybrid meta‑heuristic that leverages fuzzy logic to enhance Differential Evolution for grid job scheduling. The experimental results substantiate the claim that the proposed Fuzzy‑DE delivers more optimal schedules and faster convergence than several leading alternatives, while also identifying clear avenues for further refinement and broader applicability.