Lock-Free Collaboration Support for Cloud Storage Services with Operation Inference
summary
Reverse engineering and measurements reveal long synchronization tails, missing locks, and many false conflicts in popular cloud drives. The system infers per-file edit sequences using an edit-graph LCS-to-shortest-path trick (huge speedup over O(N^2)), transforms and merges operations to preserve intent, and implements this as UFC2 on AWS. Evaluation shows 98% conflicts avoided with about 2% added upload time and lower network cost versus full-file sync, while true semantic clashes still require manual resolution.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:44Collaboration with cloud storage services
- 1:49The measurement study
- 2:28Package capturing, traffic decrypting, and docs & source reading
- 3:13Major findings (1)
- 6:00Inference implicit operation sequences (1)
- 7:42Automatic detecting true conflicts
- 9:56Handling multiple conflicting versions
- 11:15Overhead in conflict resolution
- 11:50Overhead in network traffic
- 12:30In summary
- 14:22Effectiveness of conflict resolution
related talks
24mEngineering Real-Time Data Synchronization Across Multi-Cloud Systems
15mParaSync: Exploiting Fine-Grained Parallelism for Efficient File Synchronization
28mAsynchronous Directory Operations in CephFS
21mNew Techniques to Improve Small I/O Workloads in Distributed File Systems
45mAutomerge: A new foundation for collaboration software
15mUsing Dynamically Layered Definite Releases for Verifying the RefFS File System
44mOpenSearch’s Quest for Disaggregated Storage: The Strangler Fig Patter...
20mCalcspar: A Contract-Aware LSM Store for Cloud Storage with Low Latency Spikes