FTI-TMR: A Fault Tolerance and Isolation Algorithm for Interconnected Multicore Systems

TL;DR

FTI-TMR tackles the energy cost and limited fault tolerance of traditional TMR by introducing a stability-score driven, distributed fault-detection framework for interconnected multicore systems. It eliminates reliance on hardware-based fault detectors by electing two highly stable nodes to periodically diagnose and isolate permanent faults across the network, leveraging a Raft-inspired leader model and ABFT-style coordination. A formal stability metric $SS_i = \frac{S_i - D_i + f_i}{S_i + F_i}$ guides leader selection and fault isolation, achieving ~30% lower task workload than baseline TMR and higher permanent fault coverage than Reactive TMR, as shown in simulation with STG workloads and a nine-node, 4-core setup. This approach advances reliable, energy-efficient fault tolerance for large-scale multicore deployments and could extend to distributed autonomous systems where reliable peer diagnosis is desired.

Abstract

Two-Phase TMR conserves energy by partitioning redundancy operations into two stages and making the execution of the third task copy optional, yet it remains susceptible to permanent faults. Reactive-TMR (R-TMR) counters this by isolating faulty cores, handling both transient and permanent faults. However, the lightweight hardware required by R-TMR not only increases complexity but also becomes a single point of failure itself. To bypass isolated node constraints, this paper proposes a Fault Tolerance and Isolation TMR (FTI-TMR) algorithm for interconnected multicore systems. By constructing a stability metric to identify the most reliable nodes in the system, which then perform periodic diagnostics to isolate permanent faults. Experimental results show that FTI-TMR reduces task workload by approximately 30% compared with baseline TMR while achieving higher permanent fault coverage.

Figures (9)

• Figure 1: TMR schemes: (a) Conventional TMR: three task copies must be executed. (b) Two-Phase TMR: the third copy can be omitted under normal conditions. (c) Enhanced Two-Phase TMR: task copies are distributed across different cores, allowing tolerance to permanent core faults. (d) Reactive TMR: based on Enhanced Two-Phase TMR, a permanent-fault detector is introduced to disable the use of faulty cores according to execution history.
• Figure 2: Raft consensus algorithm.
• Figure 3: Limitations of R-TMR and Improvement Concept: (a) the system fails when there is fewer than 2 fault-free cores, or (b) when the additional fault-detection hardware becomes defective. Improvement Concept: (c) the two most stable nodes are selected to perform permanent fault detection on the other nodes.
• Figure 4: Primary leader election.
• Figure 5: Secondary leader election.