@article{ART003353362},
author={Baek Young Mi and Park jung-kyu},
title={Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B},
journal={Journal of Internet of Things and Convergence},
issn={2466-0078},
year={2026},
volume={12},
number={3},
pages={14}
TY - JOUR
AU - Baek Young Mi
AU - Park jung-kyu
TI - Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B
JO - Journal of Internet of Things and Convergence
PY - 2026
VL - 12
IS - 3
PB - The Korea Internet of Things Society
SP - 14
EP -
SN - 2466-0078
AB - This study examines how Linux filesystem choice affects the startup performance of Gemma 4 E2B in a local inference environment. Using a fixed hardware platform, we compared ext4, XFS, and Btrfs under cold-cache and warm-cache conditions, and repeated each experiment five times under the same model, prompt, and runtime settings. The analysis focused on wall time and load duration, while additional internal metrics were collected to identify which stage contributed most to the observed delay. Under cold-cache conditions, the three filesystems showed clear performance differences, and Btrfs achieved the lowest wall time and load duration in the tested setup. In contrast, the performance gap became much smaller under warm-cache conditions. The results also show that the change in end-to-end latency was driven more by the model-loading stage than by the generation stage itself. These findings indicate that startup behavior in local LLM deployment is influenced not only by compute capability but also by filesystem choice and page-cache state. The study suggests that storage configuration should be considered together with workload characteristics when designing practical local AI systems.
KW - Gemma 4 E2B;Local LLM Inference;Linux Filesystems;Startup Performance;Page Cache
DO -
UR -
ER -
Baek Young Mi and Park jung-kyu. (2026). Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B. Journal of Internet of Things and Convergence, 12(3), 14.
Baek Young Mi and Park jung-kyu. 2026, "Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B", Journal of Internet of Things and Convergence, vol.12, no.3 14.
Baek Young Mi, Park jung-kyu "Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B" Journal of Internet of Things and Convergence 12.3 14 (2026) : 14.
Baek Young Mi, Park jung-kyu. Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B. 2026; 12(3), 14.
Baek Young Mi and Park jung-kyu. "Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B" Journal of Internet of Things and Convergence 12, no.3 (2026) : 14.
Baek Young Mi; Park jung-kyu. Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B. Journal of Internet of Things and Convergence, 12(3), 14.
Baek Young Mi; Park jung-kyu. Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B. Journal of Internet of Things and Convergence. 2026; 12(3) 14.
Baek Young Mi, Park jung-kyu. Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B. 2026; 12(3), 14.
Baek Young Mi and Park jung-kyu. "Effects of Linux Filesystems on the Startup Performance of Local Inference with Gemma 4 E2B" Journal of Internet of Things and Convergence 12, no.3 (2026) : 14.
