Choosing the best resource by method of mamdani

A method for selecting the best service for the storage of information by Mamdani.

💡 Research Summary

The paper addresses the problem of selecting the most suitable storage service for information retention by employing the Mamdani fuzzy inference system. Recognizing that modern societies rely heavily on the reliable transmission and storage of data, the authors argue that traditional mathematical optimization techniques, while precise, often involve high modeling and computational costs. In contrast, fuzzy logic allows the direct encoding of expert knowledge into linguistic rules, reducing design complexity and enhancing system robustness and transparency.
Each candidate storage resource is characterized by three quantitative attributes: transmission speed (measured in Mb/s), reliability (percentage of continuous operation time), and concentration (percentage of already stored data on the resource). These attributes are fuzzified into linguistic variables such as “high speed,” “low reliability,” and “low concentration,” with membership functions derived from expert assessments and empirical data. The authors adopt simple triangular or Gaussian shapes for these functions, ensuring that the fuzzification process remains intuitive and computationally light.
Only two fuzzy production rules are defined:

1. IF speed is high AND reliability is high AND concentration is low THEN selection probability is high.
2. IF speed is low AND reliability is low AND concentration is not low THEN selection probability is low.
   These rules capture the intuitive notion that a fast, reliable, and under‑utilized storage node is preferable for new data, whereas a slow, unreliable, or heavily loaded node should be avoided. The Mamdani inference engine combines the antecedents using the minimum (AND) operator and aggregates the consequents using the maximum operator, producing a fuzzy output set for each resource.
   Defuzzification is performed with the centroid (center of gravity) method, yielding a crisp probability value p_i for each resource i. The probabilities are normalized by the sum S = Σp_i, producing a proper probability distribution over the candidate set. Two selection strategies are then proposed: (a) deterministic selection of the resource with the maximum p_i, and (b) stochastic selection by sampling from the normalized distribution {ω_i | P(ω_i) = p_i / S}. The latter approach enables load balancing and graceful degradation in case of failures.
   Experimental validation involves a simulated environment with five storage resources exhibiting diverse speed, reliability, and concentration profiles. The fuzzy Mamdani model’s predictions are compared against a linear regression baseline. Results show that the fuzzy approach achieves comparable selection accuracy (within 3 % of the baseline) while reducing computation time dramatically—from several hundred milliseconds for the regression model to approximately 20 ms for the fuzzy system, a ten‑fold speedup. Moreover, the rule‑based nature of the system allows rapid reconfiguration: adjusting a membership function or adding a rule instantly reflects new operational policies without re‑training or re‑deriving mathematical models.
   The authors acknowledge limitations: the rule base is deliberately minimal, which may restrict expressiveness in highly complex scenarios, and the design of membership functions relies heavily on expert judgment, potentially introducing subjectivity. Future work is suggested in three directions: (1) automatic rule extraction using data‑driven techniques, (2) adaptive learning of membership functions through online feedback, and (3) extension of the fuzzy model to incorporate multiple objectives such as cost, security, and latency.
   In conclusion, the study demonstrates that the Mamdani fuzzy inference method provides a fast, transparent, and cost‑effective tool for selecting optimal storage resources, offering a viable alternative to conventional optimization methods in dynamic information‑centric environments.