Efficient protection for routing programs via semantics-aware instruction vulnerability assessment

Abstract

High-energy particles can silently corrupt routing computations in space-based networks, yet conventional failure signals may not expose the resulting silent data corruption (SDC). Selective instruction duplication mitigates SDC, but frequency-based selection ignores routing semantics, and methods based on fault injection (FI) are extremely time-consuming. We propose RSVAP, a compiler-level framework that builds a program graph from instruction execution frequencies and routing semantics, including path cost, predecessor state, and reconstructed path output. The graph helps identify vulnerable instructions that can corrupt routing results. Based on a small number of valid fault-injection trials, RSVAP then uses a 0-1 knapsack algorithm to select a protection set under a fixed cost budget. On MiBench Dijkstra, it reduces the unprotected SDC rate from 2.78% to 0.36%, 0.24%, and 0.06% under 30%, 50%, and 70% dynamic-cost budgets. RSVAP approaches the protection effectiveness of FI-Guided SID while reducing learning injections by 90%, and achieves the same observed SDC count at the 70% budget.