On Scalable Integrity Checking for Secure Cloud Disks
summary
Paper quantifies storage-level Merkle tree costs and shows hashing is bottleneck, then designs Dynamic Merkle Trees that bias hot leaves toward the root. Implementation is a block device driver (~5,000 lines of C++) running on Linux and evaluated on EC2 with local NVMe. DMTs learn workload locality, reduce hash work, achieve about 85% of optimal throughput, report up to 2.2× throughput/latency gains and ~40% tail-latency improvement.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:19Cost of integrity protection
- 3:06Cryptographic proof systems
- 5:55Reference locality in hash trees
- 9:51Tree oracle performance
- 10:56Dynamic Merkel trees
- 13:48Implementation setup
- 14:33Scalability with DMTs
- 16:21Adaptability to workloads
- 17:04Summary
related talks
13mMlsDisk: Trusted Block Storage for TEEs Based on Layered Secure Logging
14mNimble: Rollback Protection for Confidential Cloud Services
41mTPMs, Merkle Trees and TEEs: Enhancing SLSA with Hardware-Assisted...
27mSecure Distributed Block Device for Edge and Cloud
48mDirty Tricks of Computer Hardware
01h 01mKey transparency: Blockchain meets NoiseSocket
13mA Log-Structured Merge Tree-aware Message Authentication Scheme for Persistent...
01h 03mHow I Learned to Stop Worrying and Trust the Cloud