Speculative Encryption on GPU Applied to Cryptographic File Systems
summary
Implements EncFS++ with a CUDA/GPU backend (wise/wiseT) and libfuse, using CTR mode to speculatively generate encryption masks on the GPU and a per-volume global counter plus on-disk nonce nodes (128-bit nonces per block) to meet CTR uniqueness. A virtual circular context pool prioritizes mask production for sequential reads/writes. Tests on a 16 GiB file and filebench workloads show significant throughput and CPU-efficiency gains, GPU active only ~12–20%, and usable speedups even for 4 KiB blocks.
AI-written summary · about
chapters (5)
- 0:00Start
- 8:03Gpu Encryption Acceleration
- 20:02Macro Benchmarks
- 21:36Conclusions
- 23:42Why Did You Not Test against Counter Mode on Cpu
related talks
33mBringing fscrypt Encryption to CephFS Userspace
33mEncryption for filesystems with advanced features new fscrypt functionality
56mBuilding a compiler and processor for fast processing of encrypted data
20mTrusted Hardware for AI Workloads: Extending Confidential...
26mWide-Block Cipher Support and HCTR2
30mEncrypting Memory at Scale
16mAutomatically Eliminating Speculative Leaks from Cryptographic Code with Blade
53mSecure Embedded Linux Product - A Success Story