[Paper Review] Fault Masking By Probabilistic Voting

Barış Baykant Alagöz|ArXiv.org|Jan 9, 2009

Software Testing and Debugging Techniques3 references3 citations

TL;DR

This paper proposes a probabilistic voter for Multi-Modular Redundancy (MMR) systems that improves fault masking by incorporating module functionality and symbol probabilities into voting decisions, outperforming conventional majority voting in availability under transient errors, as demonstrated through simulations with 3 and 5 redundant modules.

ABSTRACT

In this study, we introduced a probabilistic voter, regarding symbol probabilities in decision process besides majority consensus. Conventional majority voter is independent of functionality of redundant modules. In our study, proposed probabilistic voter is designed corresponding to functionality of the redundant module. We tested probabilistic voter for 3 and 5 redundant modules with random transient errors inserted the wires and it was seen from simulation results that Multi-Modular Redundancy (M-MR) with Probabilistic Voting (PV) had been shown better availability performance than conventional majority voter.

Motivation & Objective

• To address the limitation of conventional majority voters, which ignore module functionality and rely solely on consensus.
• To improve system availability in the presence of transient errors by leveraging probabilistic decision-making based on module reliability.
• To design a voting mechanism that adapts to the functional state and output probabilities of redundant modules.
• To evaluate the performance of the proposed probabilistic voter against traditional majority voting in MMR architectures.

Proposed method

• The probabilistic voter uses symbol probabilities from redundant modules as inputs to the voting process, rather than treating all modules equally.
• Module functionality is assessed dynamically, and voting weights are adjusted based on the likelihood of correct output per module.
• The voter computes a weighted vote using probabilities derived from module behavior, favoring more reliable modules.
• Simulations were conducted with 3 and 5 redundant modules, injecting random transient errors into wires to test fault masking capability.
• The system uses a probabilistic voting function that combines outputs based on statistical confidence in each module’s correctness.
• Performance is measured by system availability, defined as the probability of correct output under error conditions.

Experimental results

Research questions

• RQ1 How does incorporating module functionality and symbol probabilities into voting affect fault masking in MMR systems?
• RQ2 Can a probabilistic voter outperform conventional majority voting in terms of system availability under transient errors?
• RQ3 What is the impact of varying the number of redundant modules (3 vs. 5) on the effectiveness of probabilistic voting?
• RQ4 How does the dynamic weighting of module outputs influence system resilience to transient faults?

Key findings

• The proposed probabilistic voter demonstrated superior availability performance compared to conventional majority voting in MMR systems.
• With 3 redundant modules, the probabilistic voter achieved higher fault masking capability under random transient error injection.
• With 5 redundant modules, the improvement in availability was more pronounced, indicating scalability of the approach.
• The integration of module functionality and symbol probabilities significantly enhanced decision accuracy during error conditions.
• Simulation results confirmed that probabilistic voting reduces the likelihood of incorrect output decisions under fault scenarios.
• The method effectively leverages reliability information from redundant modules to improve overall system resilience.