Calcspar: A Contract-Aware LSM Store for Cloud Storage with Low Latency Spikes
summary
Calcspar is an LSM-store variant built to tame Amazon EBS latency by regulating I/O rate and absorbing bursts with a fluctuation-aware cache and a congestion-aware IOPS allocator. Implemented on RocksDB and tested on EBS with real-world workloads, it keeps 99th-percentile reads under 550 μs, cuts average latency by 66%, and preserves normal read/write throughput while reducing tail spikes compared to state-of-the-art LSM stores.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:54Motor and Latency Performance
- 5:11Latency Specs
- 9:05Applications
- 10:47Application Example
- 12:35Challenges
- 14:35Solution
- 18:12Conclusion
related talks
32mLow Latency Serving of Offline Data: Efficient, Safe, and Reliable Data...
23mRobustness in the Salus Scalable Block Store
35mSystems @Scale Summer Q&A Week 3
24mStronger Semantics for Low-Latency Geo-Replicated Storage
47mGoogle Cloud Spanner: Global consistency at scale
16mHolistic and Automated Task Scheduling for Distributed LSM-tree-based Storage
19mScalog: Seamless Reconfiguration and Total Order in a Scalable Shared Log
33mGoogle Cloud Spanner: Global consistency at scale