ScaleFlux CSware– unlocking the full potential of your ScaleFlux CSDs, SSDs, and CSPs
SFX 3000 Storage Processor​
NOV 04 / 2021 PDF
NSD 3000 NVMe SSDs
NOV 04 / 2021 PDF
CSD 3000 NVMe Computational Storage Drives​
NOV 04 / 2021 PDF


With CSware, ScaleFlux is making it easier for you to get more from your ScaleFlux storage solutions.    ​

CSware includes not only complete applications, but also enhancements to existing applications, integrations into applications, plug-ins, and example code.  ​

For the initial wave of CSware, ScaleFlux has 3 projects in development:​

  • KallaxDB – a KV Store optimized for computational storage drives​

  • CSware RAID – software RAID that takes full advantage of transparent compression in the drives​

  • B-tree optimizations – modifications to the B-tree to improve random write performance in relational databases 


Multiply KV Store Efficiency

KV stores, such as RocksDB, are commonly used to handle large volumes of unstructured data, particularly when the individual data items are small (e.g. <256 Bytes).  The data structures in KV Stores can be quite complex, consuming large amounts of memory and high amounts of CPU processing power. ​

Leveraging in-drive transparent compression, KallaxDB employs a simple data structure to largely reduce the indexing CPU/memory overhead. ​

The results:  ​

  • Higher Efficiency​

  • Lower CPU burden per operation​

  • Less DRAM needed per TB of data in the data set​

  • Better Performance​

  • Higher operations per second​

  • Lower latency per operation



Multiply Relational Database Performance​

Relational databases, such as MySQL, PostgreSQL, MariaDB, Oracle, SQL Server, and MongoDB, use B-tree data structures.  B-trees are great for fast read performance.  However, they can suffer from low write performance because random updates/inserts could cause very high B-tree write amplification.​

ScaleFlux has demonstrated the feasibility for B-tree to take advantage of in-drive transparent compression to largely reduce the write amplification, leading to much higher B-tree write performance​

The results:​

  • 10x reduction in write amplification​

  • Improved write performance & latency​

  • Maintains read performance & storage efficiency​




Multiply Software RAID Efficiency

With the broad adoption of NVMe SSDs for high-performance applications, users must balance the costs of enabling fault tolerance in their NVMe Flash tier with performance.  ​

A RAID 5 configuration saves on the total amount of storage capacity needed to store a user’s data.  But, it carries significant overhead for parity calculations and requires more time to rebuild data.​

RAID 10 gets rid of the parity calculation overhead and improves rebuild performance.  But, it significantly increases the storage costs by requiring 2TB of drive capacity per 1TB of user data.​

ScaleFlux CS RAID can give users the higher performance of RAID 10 with the lower storage capacity costs of RAID 5.  ScaleFlux CS RAID monitors the compression ratio of the data and the physical free space on the drives.  When there is sufficient compressibility in the data, the SW uses the excess free space on the drives to perform like RAID 10… without increasing the physical number of drives or physical capacity per drive.​

The results:​

  • Reduced CPU burden for parity calculations​

  • RAID 5 storage cost and capacity structure​

  • RAID 10 performance and rebuild times



Comparing SW RAID Solutions