More Optimizations Has Linux Approaching 7M IOPS Per Core

Linux block subsystem maintainer and IO_uring lead developer Jens Axboe continues making staggering optimizations to the kernel code to squeeze out the maximum performance potential out of his shiny new system.

After upgrading to an AMD Ryzen 9 5950X system for development and keeping to his Intel Optane Gen2 storage, he ended September being able to achieve around 5.1M IOPS per-core for this system he uses for reference testing and evaluation of Linux block subsystem changes. That 5.1M IOPS per-core is impressive while previously he was happy hitting above 3M IOPS per-core.

Now though with just a little more than a week into pressing his new system, he has already surpassed 6M IOPS with work-in-progress kernel changes and is now quickly approaching 7M IOPS per-core on the same platform.

Updated the branch, see link in comment for tweet below. The only way is up:

IOPS=6837504, BW=3338MiB/s, IOS/call=32/31, inflight=(73 69)
IOPS=6823296, BW=3331MiB/s, IOS/call=32/32, inflight=(110 46)

Hopefully we'll break 7M next week. https://t.co/wZiCgpfEZG

— Jens Axboe (@axboe) October 9, 2021

Axboe shared that he's up to around 6.8M IOPS per-core now as he ended out last week. He noted that this week he hopes to "break 7M" as his latest self-imposed challenge for this Linux engineer currently employed by Facebook.

His patches pushing the greater performance have been changes to the block code, NVMe, multi-queue blk-mq, and IO_uring. His improvements are being staged via linux-block's "perf-wip" branch. Hopefully most of these patches at least will be ready for mainlining in the 5.16 merge window in November.