Architecture Dependent Temporal Observability Under Deployment Interference in Edge Inference Systems

TL;DR

This paper demonstrates that deployment interference can corrupt both inference timing and its measurement infrastructure in edge systems, revealing failure modes hidden by software-only timing analysis.

Contribution

It empirically shows that external timing observability can fail independently of software reports, affecting the accuracy of latency measurements in edge inference systems.

Findings

External timing measurements reveal failures not visible in software logs.

Different inference architectures exhibit distinct latency distribution patterns.

Storage stress can cause complete timing measurement failures despite normal software logs.

Abstract

Edge inference systems are typically evaluated with software-reported latency collected under controlled conditions. We argue, and demonstrate empirically, that deployment interference can corrupt not only the inference timing being measured but the timing observability infrastructure that measures it, and that the two failures can occur independently. We pair software-reported timing with externally observable GPIO intervals captured by a Saleae Logic Pro 8 logic analyzer on an NVIDIA Jetson Orin Nano, running MobileNetV2 under two inference architectures (TensorRT FP16 GPU and ONNX Runtime CPU) across baseline, light memory pressure, and storage writeback stress. Across 35 paired capture runs (3500 samples) plus 3 storage-stress runs where external pairing failed (300 software-only samples), we observe three findings the software-only view does not surface. (1) The two architectures differ not only in mean latency but in distributional structure: TensorRT baseline clusters tightly near 1.23 ms (run-mean SD 15 us) while ORT CPU baseline is multimodal with run-mean SD 31.8 ms. (2) Light memory pressure inflates TensorRT P99 from 1.28 ms to 1.61 ms, while one of five ORT memory-stress runs collapses into a deterministic 198 ms regime rather than uniformly inflating variance. (3) All three TensorRT storage-stress runs produce complete software timing logs (100/100 iterations) alongside externally observable timing failures of three different kinds (full post-marker collapse, ~40% transition loss, and complete acquisition failure) -- while the runtime reports normal completion in every case. We claim, narrowly, that timing observability is itself an interference-sensitive resource, and that summary statistics from a single timing source can hide failure modes an independent external observer makes visible.