Linux 7.3 Improving RAID 5/6 Benchmark-Based Algorithm Selection

Following the main set of memory management "MM" changes submitted for Linux 7.3, Andrew Morton sent in his "non-MM" pull request of the other patches he's been collecting from various developers for this next Linux kernel release. Standing out there were some improvements to the RAID 5/6 benchmark-based algorithm selection.

Linux selects its optimal RAID XOR engine (used by RAID5 and RAID6 software configurations) and the RAID6 specific code paths based on an internal speed test during the kernel module initialization. This is done for picking the fastest hardware-accelerated method available for the given hardware such as for leveraging AVX2 or AVX-512. With the Linux 7.3 kernel, various improvements are made to the benchmark-based algorithm selection for the XOR and RAID6.

In addition to various improvements around the benchmark-based algorithm selection at boot time, there is also now KUnit benchmark coverage for the XOR and RAID6 code plus other code improvements. Among the code changes are using 4 + 1 disks for a more realistic load calculation and reporting the throughput on the data disks rather than just that of the parity disk. No data was provided on the patches to quantify the impact or further highlight the benefits, but any enhancements ultimately are welcome.

That was the main takeaway for me from the non-MM pull request for the Linux 7.3 kernel with much of the other material amounting to routine bug fixes.