Intel Database Reference Stack 1.0 Shows How To Optimize For Optane DC Persistent Memory

Similar to their Deep Learning Reference Stack for delivering better AI / deep learning performance on Xeon Scalable CPUs, Intel has now released the Database Reference Stack 1.0 for optimizing performance on their DC Persistent Memory.

Intel's Database Reference Stack 1.0 is another initiative under their Clear Linux umbrella for demonstrating a database setup that is optimized for their Optane DC Persistent Memory hardware in conjunction with 2nd Gen Xeon Scalable (Cascadelake) CPUs.

This early release of their Database Reference Stack runs atop Clear Linux while tailoring Apache Cassandra and Redis for running off these persistent memory modules. This stack also pulls in Kubernetes for orchestration/management, Kata Containers, and makes use of the Intel Persistent Memory Development Kit (PMDK) and Low-Level Persistence Library. Both FSDAX and DEVDAX persistent memory modes are supported by this stack.

Compared to running Apache Cassandra off enterprise NVMe SSDs, moving to Intel Optane DC Persistent Memory in the app direct mode can yield throughput improvements as much as 6.5~8.6x faster with the Database Reference Stack. But the benefits of the Database Reference Stack 1.0 is obviously much less compared to a conventional software stack more simply deployed on the DCPMM.

Those wanting to learn more about Intel's Database Reference Stack 1.0 release can do so at ClearLinux.org.