Micron Announces An Open-Source Storage Engine Designed For SSDs, Persistent Memory
Micron announced today the "World’s First Open-Source Storage Engine Designed for SSDs and Storage Class Memory"... Or simply put, yet another key-value store database and this time designed for high performance SSDs and persistent memory.
Micron noted in today's press release, "HSE improves throughput of particular storage applications by up to six times, reduces latency 11 times and improves SSD endurance by seven times. HSE can also take advantage of multiple classes of media concurrently, such as flash and 3D XPoint technology. When a Micron X100 NVMe SSD, the world’s fastest SSD, is added to a set of four Micron 5210 QLC SSDs, throughput more than doubles and read latency improves nearly four times."
The Heterogeneous-Memory Storage Engine (HSE) itself is a key-value store and under the project umbrella is also a MongoDB implementation.
This HSE storage engine depends upon an mpool object storage media pool that is implemented as a Linux kernel module. Mpool interfaces with SSDs or other persistent memory storage directly, bypassing file-systems and other overhead while also supporting replication across classes of memory and other storage features. Mpool itself appears to be usable outside of HSE albeit so far is the only known user of this new Micron kernel module. Until (if) this Mpool kernel module is upstreamed to the Linux kernel will be a major blocker for adoption.
Micron advertises HSE as being ideal for NoSQL, SDS, HPC, Big Data, IoT, and AI solutions. Most of the public performance numbers for HSE are putting it up against the Facebook-backed RocksDB database and MongoDB.
The code to HSE can be found via GitHub.
I'll be working on some HSE benchmarks shortly.
Micron noted in today's press release, "HSE improves throughput of particular storage applications by up to six times, reduces latency 11 times and improves SSD endurance by seven times. HSE can also take advantage of multiple classes of media concurrently, such as flash and 3D XPoint technology. When a Micron X100 NVMe SSD, the world’s fastest SSD, is added to a set of four Micron 5210 QLC SSDs, throughput more than doubles and read latency improves nearly four times."
The Heterogeneous-Memory Storage Engine (HSE) itself is a key-value store and under the project umbrella is also a MongoDB implementation.
This HSE storage engine depends upon an mpool object storage media pool that is implemented as a Linux kernel module. Mpool interfaces with SSDs or other persistent memory storage directly, bypassing file-systems and other overhead while also supporting replication across classes of memory and other storage features. Mpool itself appears to be usable outside of HSE albeit so far is the only known user of this new Micron kernel module. Until (if) this Mpool kernel module is upstreamed to the Linux kernel will be a major blocker for adoption.
Micron advertises HSE as being ideal for NoSQL, SDS, HPC, Big Data, IoT, and AI solutions. Most of the public performance numbers for HSE are putting it up against the Facebook-backed RocksDB database and MongoDB.
The code to HSE can be found via GitHub.
I'll be working on some HSE benchmarks shortly.
3 Comments