SimEnc: A High-Performance Similarity-Preserving Encryption Approach for...
summary
SimEnc builds a prototype in Go and C++ with one GPU and integrates semantic-hash MLE to make similar chunks yield the same encryption key via adaptive clustering, plus a Huffman partial decoding space to avoid full LZ77 recompression. A Bloom-filter header check picks which layers need chunk-level work. Evaluated on Docker Hub and IBM workloads, it raises deduplication ratio ~38–109% versus prior MLE methods, cuts pull latency up to 72.4%, and claims storage savings up to 261.7% and 54.2%.
AI-written summary · about
chapters (13)
- 0:00SimEnc introduction
- 1:06Deduplicating encrypted images
- 2:54MLE-based approach limits
- 4:40Image layer decompression
- 6:03Partial decoding for deduplication
- 8:10SimEnc key contributions
- 8:52System architecture
- 9:56Deduplication modes trade-offs
- 11:01Semantic hashing
- 13:30Implementation setup
- 14:12Performance vs baselines
- 15:58Latency and throughput
- 17:01Conclusion and future work
related talks
20mSecure and Lightweight Deduplicated Storage via Shielded Deduplication...
33mHomomorphic Encryption in the Open Source World
37mOpen-source Machine Learning on Encrypted Data
15mAccelerating Encrypted Deduplication via SGX
41mNon-traditional Encryption Problems And Solutions
40mConfidential Computing and Privacy-Enhancing Technologies - The Landscape
41mTechnical Evaluation of a Disruptive Approach in
35mEncrypted Computation: What if Decryption Wasn’t Needed?