A Per-Access Upper Bound for Shared-Resource Interference in Direct-Mapped Multicore Architectures
Provides a formal, traceable interference bound for safety-critical multicore systems under DO-178C/CAST-32A certification, but is limited to specific architectural constraints.
The paper proves a per-access upper bound of (N-1)Lmem for shared-resource interference in direct-mapped multicore processors under strict invariants, enabling traceable separation of interference from WCET budgets for certification.
We present a formal bounding analysis for maximum credible interference in multicore processors under strict architectural invariants: direct-mapped L2 cache (1-way associativity), disabled Miss Status Handling Registers (MSHRs), single-bank main memory, deterministic pinned tasks with fixed physical memory mapping, and a pessimistic L2/memory arbitration policy. We prove that, under these invariants, the per-critical-access stall imposed on a target task T is bounded above by (N-1)Lmem, and that this bound is attained by a synchronized adversarial workload of N-1 congruent-different-tag memory requests issued in phase with T's critical accesses. The argument is per-access and direct, requiring no informal multiplicative interference function. The derivation is purely analytical and discussed in the context of DO-178C/CAST-32A certification objectives for airborne software. Limitations and conditions for applicability are explicitly stated. This work provides a traceable method for separating multicore interference from Worst-Case Execution Time (WCET) budgets under fixed architectural constraints.