@article{ART003399396},
author={Hyun-Seob Lee},
title={Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG},
journal={Journal of Internet of Things and Convergence},
issn={2466-0078},
year={2026},
volume={12},
number={5},
pages={9}
TY - JOUR
AU - Hyun-Seob Lee
TI - Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG
JO - Journal of Internet of Things and Convergence
PY - 2026
VL - 12
IS - 5
PB - The Korea Internet of Things Society
SP - 9
EP -
SN - 2466-0078
AB - Retrieval-Augmented Generation (RAG) systems based on large language model (LLM) convert large volumes of documents into embedding vectors, store them in a vector database, and repeatedly perform similarity search and update operations for each query. Understanding the storage-level access patterns induced by these vector operations is essential for improving system performance and optimizing storage resources. However, traces from recent AI applications such as RAG are rarely publicly available, posing a significant obstacle to related storage system research. Prior storage research has largely relied on synthetic traces—typically random or sequential access patterns—to evaluate storage performance. However, this approach fails to adequately capture the vector access characteristics unique to real AI/RAG applications, such as similarity-based irregular access and repetitive search-and-update patterns. This study designs and implements a Trace Generation Framework that produces storage access traces derived from actual RAG operations. We construct a knowledge base from actual Wikipedia documents, split the documents into chunks, and generate embedding vectors for each chunk, which are then stored in a FAISS-based vector database. Vector access patterns are obtained by repeatedly performing search operations that emulate real RAG queries, and each vector access is mapped to a Logical Block Address (LBA) on a NAND flash-based storage device to generate a storage access trace. The traces generated by this framework are expected to serve as foundational data for future research on storage performance evaluation and resource optimization for AI applications, such as RAG.
KW - RAG;Address Pattern;Pattern Analysis;Data Analysis;Address Generator
DO -
UR -
ER -
Hyun-Seob Lee. (2026). Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG. Journal of Internet of Things and Convergence, 12(5), 9.
Hyun-Seob Lee. 2026, "Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG", Journal of Internet of Things and Convergence, vol.12, no.5 9.
Hyun-Seob Lee "Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG" Journal of Internet of Things and Convergence 12.5 9 (2026) : 9.
Hyun-Seob Lee. Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG. 2026; 12(5), 9.
Hyun-Seob Lee. "Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG" Journal of Internet of Things and Convergence 12, no.5 (2026) : 9.
Hyun-Seob Lee. Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG. Journal of Internet of Things and Convergence, 12(5), 9.
Hyun-Seob Lee. Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG. Journal of Internet of Things and Convergence. 2026; 12(5) 9.
Hyun-Seob Lee. Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG. 2026; 12(5), 9.
Hyun-Seob Lee. "Design and Implementation of a Trace Generation Framework for Storage Systems based on RAG" Journal of Internet of Things and Convergence 12, no.5 (2026) : 9.
