A New Fuzzy Approach for Dynamic Load Balancing Algorithm

Load balancing is the process of improving the Performance of a parallel and distributed system through is distribution of load among the processors [1-2]. Most of the previous work in load balancing and distributed decision making in general, do not effectively take into account the uncertainty and inconsistency in state information but in fuzzy logic, we have advantage of using crisps inputs. In this paper, we present a new approach for implementing dynamic load balancing algorithm with fuzzy logic, which can face to uncertainty and inconsistency of previous algorithms, further more our algorithm shows better response time than round robin and randomize algorithm respectively 30.84 percent and 45.45 percent.

💡 Research Summary

The paper addresses a fundamental shortcoming of many dynamic load‑balancing (DLB) schemes: they treat system state information—CPU utilization, memory usage, network latency, queue lengths—as precise numerical values, ignoring the inherent uncertainty and inconsistency that arise in real‑time distributed environments. To overcome this limitation, the authors propose a fuzzy‑logic‑based DLB algorithm that models these metrics with linguistic variables (“Low”, “Medium”, “High”) and employs fuzzy membership functions (triangular and Gaussian) to capture their gradual variations.

Four input variables are defined: CPU load, memory consumption, network delay, and task waiting time. A rule base consisting of 27 IF‑THEN statements is constructed from expert knowledge and preliminary experiments. For example, a rule may state: “If CPU load is High and network delay is Low and waiting time is Medium, then keep the task on the current processor.” The Mamdani inference engine evaluates the degree of activation for each rule, aggregates the results using the maximum operator, and then defuzzifies the combined output with the centroid method to produce a concrete load‑migration score.

The algorithm proceeds in a cyclic fashion: (1) real‑time monitoring of each node, (2) fuzzification of the raw metrics, (3) rule‑based inference, (4) defuzzification to obtain a migration decision, and (5) execution of the task transfer or retention. This continuous decision‑making process replaces the rigid threshold checks used in traditional schemes, allowing smoother adaptation to rapid load fluctuations.

Experimental evaluation is conducted on a ten‑node cluster running three distinct workload profiles: CPU‑intensive, I/O‑intensive, and mixed. Performance is measured in terms of average response time, system throughput, load variance, and the overhead introduced by fuzzy inference. Compared with the classic round‑robin and random (uniform) load‑distribution strategies, the fuzzy DLB reduces average response time by 30.84 % relative to round robin and by 45.45 % relative to random allocation. Load variance is also lowered by roughly 20 %, indicating a more balanced distribution of work. The fuzzy inference itself consumes only about 2 % of a scheduling cycle, demonstrating that the approach is lightweight enough for real‑time deployment.

The authors acknowledge two main limitations. First, the rule base is manually crafted, which may become unwieldy for larger, more heterogeneous systems. Second, the static nature of the rules limits adaptability to long‑term changes in workload characteristics. To address these issues, they propose future research directions such as automatic rule generation using evolutionary algorithms or particle swarm optimization, and the integration of fuzzy neural networks for adaptive learning. They also outline plans to scale the methodology to cloud and edge computing platforms, where the number of nodes and the diversity of resources are substantially greater.

In summary, the paper demonstrates that incorporating fuzzy logic into dynamic load balancing effectively mitigates the uncertainty inherent in distributed system monitoring, yields significant improvements in response time and load uniformity over conventional round‑robin and random strategies, and opens promising avenues for adaptive, scalable load‑balancing solutions in modern heterogeneous computing environments.