Improve drive life in RocksDB with CSD 3000
Offload the CPU to reduce the effects of write amplification using transparent compression
Key customer benefits
Reduce TCO
With improved SSD endurance and usable capacity.
Improve database responsiveness
Offloading storage processing to improve application latency.
Keep it simple
With no new software or drivers to install.
“ScaleFlux CSD 3000 generates
extraordinary results when paired with RocksDB
compared to an ordinary SSD.”
– Keith McKay, Senior Director of Applications Engineering
Challenge
RocksDB is an open-source, key-value (KV) storage engine optimized for fast, low-latency SSDs embedded in many workloads. While RocksDB is designed with SSDs in mind, its architecture is based on Log Structure Merge-Tree (LSM) data management. This approach results in write and read amplification as each update in RocksDB leads to cascading writes in the LSM tree, creating a performance bottleneck and excessively applying wear-and tear on the drives.
Testing showed that the drives saw write amplification of over 23 with RocksDB – for every 100GB of database size, the drive had to write 2300GB!
Solution
The ScaleFlux SSDs with transparent compression proved the easiest way to reduce the effects of application-produced write amplification with RocksDB. Using an NVMe SSD that can intelligently compress and manage data inside the drive improves drive life, increases performance, and reduces the complexity of database tuning without any additional software or application configuration.
Results
Using the ScaleFlux CSD 3000 in a RocksDB setup with a workload including random inserts and CPU-based LZ4 to compress from levels L2 and below reduced the writes to Flash by 40% while also removing the burden of compression from the CPU. This means longer drive life (and fewer service events!) with better overall database latency and performance.
Simply choose a better SSD to improve drive life and performance in RocksDB
Transparent compression for optimal results.
