Robustness in the Salus Scalable Block Store
summary
Salus implements pipeline commit to serialize commit phases while preparing writes in parallel, and an active storage replication that replaces stateless compute nodes to avoid a single corrupted forwarder wrecking all replicas. It adds end-to-end verification so clients never read corrupted blocks and tolerates arbitrary server faults short of impersonation. On a well‑provisioned cluster it cuts network transfer roughly in half and boosts aggregate write throughput by about 1.74× compared to HBase while keeping scalability similar to HBase.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:07Design
- 3:42Architecture
- 9:45Example
- 11:51Active Storage
- 14:25Evaluation
- 15:55Overhead
- 18:40Conclusion
- 19:57Questions
related talks
20mFine Grained Replicated State Machines for a Cluster Storage System
36mStreamlio: Towards a Unified End-to-end Solution
59mDistributed Abstractions
30mScaling HDFS with Consistent Reads from Standby Replicas
19mBuilding a Performant Configuration-Store Layer
24mValidity and the future of blockchains
47mA Distributed, Scalable and Highly Reliable Data Storage System
20mCalcspar: A Contract-Aware LSM Store for Cloud Storage with Low Latency Spikes