Kioxia Updates TRocksDB, and Validates Payload Sizes

Kioxia America, Inc. (formerly Toshiba Memory America, Inc.) has performed validation testing on its TRocksDB solution, which provides a blueprint to improve open source software for SSDs.

After evaluating a series of TRocksDB payload test scenarios, the company found that as the payload size increased, database performance gains continued to be realized ⁽¹⁾ with TRocksDB over` RocksDB (without the TRocksDB patch). Additionally, the firm has merged TRocksDB with the RocksDB release (version 6.4.6)  to ensure compatibility.

First introduced at FMS 2019, TRocksDB is an open source contribution that improves the life span of SSDs used with the RocksDB database, while delivering the same or improved performance (2). TRocksDB utilizes fast SSD storage while supporting I/O-bound, in-memory, write-once workloads.

“While we’re well-known for our role as a leader in SSDs and flashmemory solutions – at Kioxia, we’re alays striving to be a driving, transformational force in the storage market,” noted Eric Ries, SVP, memory storage strategy division (MSSD), Kioxia “Building on the successful rollout of TRocksDB, Kioxia continues to innovate by validating the performance gains that this solution offers over RocksDB. We are proud to contribute the TRocksDB code to the open source community to enable a wider variety of capabilities for users – and extend the overall life of their flash-based database storage.“

Flexible and designed for use with any off-the-shelf SSD, TRocksDB is an enhancement to the RocksDB database design that uses key values more efficiently with SSDs to enable improvements in storage and DRAM usage. It also reduces the repeated data rewriting ⁽²⁾ caused by application-generated write amplification ⁽³⁾. It runs on any Linux hardware supported by RocksDB. Being a member of the OSS community enables it to be part of storage innovation and industry participation that can evolve the platform to a higher level.

TRocksDB is available for download from a GitHub2 repository.

⁽¹⁾ Based on internal Kioxia testing: Comparing Payload Sizes: RocksDB vs. TRocksDB, February 2020.
⁽²⁾ Based on internal Kioxia testing: Introducing the TRocksDB Platform, December 2019.
⁽³⁾ SSD Write Amplification (WA) is a scenario that happens to SSD where data and metadata are written multiple times that increases the number of program and erase cycles. This effect can reduce the life of the SSD over time.

Read also:
FMS: Toshiba Memory America Optimizes RocksDB for SSDs
Solution improves storage endurance and utilization, reduces application write amplification by “up to 21X”.
August 5, 2019 | Press Release