본문 바로가기

Poster: Near-Data Processing for Homomorphic Ciphertext Operation Offloading over NVME-oF

문서 유형
연구출판물 · 학술대회 발표논문
등재
Scopus
피인용
Scopus 0
FWCI (Scopus)
0

저자

  1. 윤성준제1저자Pusan National University, Department of Information Convergence Engineering, Busan, South Korea
  2. 안성용Pusan National University, Department of Information Convergence Engineering, Busan, South Korea

초록

Homomorphic encryption (HE) enables direct computation on encrypted data without decryption, but suffers from significant data expansion, whereby a ciphertext can grow up to 1,000 times larger than its plaintext counterparts. This expansion creates a severe I/O bottleneck during data movement between storage and main memory for host-side computation. To address this, Near-Data Processing (NDP), a scheme for processing data near the storage unit, has emerged as one of the promising solutions to reduce the I/O-related overhead. In this paper, we leverage a Storage Performance Development Kit (SPDK)-based NDP platform to offload core operations of the CKKS HE scheme. Experimental results demonstrate that compared to the non-offloading baseline, the I/O traffic was reduced by about 99%, and the execution time was decreased by about 28%.

키워드

Big dataCloud computing securityFully homomorphic encryptionNear-data processingNVMe over Fabrics

식별자

Scopus EID
2-s2.0-105049564172