@article{ART003399401},
author={Jihun Kang},
title={Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory},
journal={Journal of Internet of Things and Convergence},
issn={2466-0078},
year={2026},
volume={12},
number={5},
pages={10}
TY - JOUR
AU - Jihun Kang
TI - Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory
JO - Journal of Internet of Things and Convergence
PY - 2026
VL - 12
IS - 5
PB - The Korea Internet of Things Society
SP - 10
EP -
SN - 2466-0078
AB - When processes load identical read-only data onto one graphics processing unit (GPU), memory use and host-to-device transfer costs increase. We propose loading registered data once through a central manager and sharing them via CUDA inter-process communication (IPC). Selective application programming interface (API) interposition handles the same cudaMemcpy() call as initial loading or reuse confirmation based on object state and process role. In matrix-vector multiplication experiments, logical common-data memory use decreased by 50% and 75% with two and four processes, respectively. Across four capacity-boundary configurations, independent loading failed with out-of-memory errors in all 20 paired trials, whereas the proposed method completed all workloads. With data resident, subsequent preparation time decreased by 95.74%, 97.93%, and 98.92% for 256 MiB, 512 MiB, and 1 GiB. Kernel performance remained comparable, with no result mismatches.
KW - GPU Memory Sharing;Multi-Process Computing;CUDA IPC;API Interposition;Data Reuse
DO -
UR -
ER -
Jihun Kang. (2026). Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory. Journal of Internet of Things and Convergence, 12(5), 10.
Jihun Kang. 2026, "Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory", Journal of Internet of Things and Convergence, vol.12, no.5 10.
Jihun Kang "Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory" Journal of Internet of Things and Convergence 12.5 10 (2026) : 10.
Jihun Kang. Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory. 2026; 12(5), 10.
Jihun Kang. "Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory" Journal of Internet of Things and Convergence 12, no.5 (2026) : 10.
Jihun Kang. Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory. Journal of Internet of Things and Convergence, 12(5), 10.
Jihun Kang. Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory. Journal of Internet of Things and Convergence. 2026; 12(5) 10.
Jihun Kang. Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory. 2026; 12(5), 10.
Jihun Kang. "Data sharing techniques between GPU processes to avoid duplicate loading of data in GPU memory" Journal of Internet of Things and Convergence 12, no.5 (2026) : 10.
