Rapid Integration
- No changes required to address translation, failure modes, consistency models or core logic
- Did not need to handle variable zone sizes or zone excursions
- No need to implement Zone Append because Relaxed Write Pointers were provided
- Did not need to account for low level NAND attributes or vendor-specific anomalies
- Did not need to perform NAND error handling or manage bad block replacement
- Configured drive without writing new mapping code
- Configured isolated regions, zones (segments) and write stripes
to best match SALSA architecture and application requirements - Did not need to account for addressing gaps inside zones or between zones because zones were presented with contiguous LBAs
- Did not have to implement wear leveling into cleaning policy or track wear metrics
- Did not have to assimilate NAND data retention requirements or implement scrubbing processes to prevent unrecoverable errors
Exceptional Performance
- Did not need to limit queue depth (no strict write pointer restriction)
- Zone Append not required so no associated latency penalty creating stripes or from mapping updates
- SALSA did not need to manage low level NAND attributes – doing this over a system bus is inefficient and introduces overhead
- Configurable zones minimized write amp., improving throughput and latency
- Configurable write stripes improved parallelism and throughput
- Configurable iso-boxes minimized collisions to improve tail latencies
- Did not have to repetitively poll device to query wear leveling and data retention metrics
Complete case study document:
Case Study: IBM Research’s integration of Radian Zoned Flash SSD with SALSA storage software stack