Unleashing The Potential of Datacenter SSDs by Taming Performance Variability
summary
Sandook is a rack-scale block storage system with a logically centralized controller plus lightweight clients and SSD agents that implements log-structured placement for writes, replica-aware read routing, read-write segregation, and TCP-like storage congestion control without special hardware. It claims to recover lost SSD capacity, unlocking ~95% of ideal throughput and delivering 30–82% higher raw I/O, 1.7× storage and 1.9× application throughput, 88% lower latency, and 12–94% end-to-end improvement across benchmarks in the paper.
AI-written summary · about
chapters (15)
- 0:00SSD performance variability
- 1:03SSD cost and data center use
- 2:07Measured performance gap in arrays
- 3:32Sources of SSD variability
- 4:55Read-write interference impact
- 5:56Garbage collection and tail latency
- 6:59Limitations of SSD routing
- 8:05Sindook’s routing design
- 9:28Time scale separation in SSD
- 10:28Sindook architecture overview
- 11:53Load steering policy
- 12:14Read-write segregation
- 15:19Performance improvements with Sindook
- 16:22Conclusion and open-source
- 17:03Q&A
