On 9/20/22 14:08, Stephen Bates wrote:
Hi fio'ers I have been working with a few others on NVMe performance in a range of different environments. At the same time I have been tracking some of the reports on IOPS per core for NVMe+io_uring [1][2]. I was hoping to create a standrd and repeatable test for this using a mix of packer, Ansible and AWS machines. The idea would be a scriptable flow to obtain a working fio+libiouring+kernel on some AWS EC2 instance(s) and then automate some tests to obtain IOPS per CPU core. To help me get started on this does anyone have any pointers on how best to configure the system to maximize this metric? Also any fio scripts for this that people can point me too would be excellent. If there is interest I would happily work this into a patch for fio (perhaps in some sort of CI way). Cheers Stephen Bates, PhD. [1]: https://www.phoronix.com/news/7.4M-IOPS-Linux-Per-Core [2]: https://twitter.com/axboe/status/1452689372395053062?lang=en
Stephen, t/one-core-peak.sh was added last year by Erwan Velu and might be useful to you:
https://github.com/axboe/fio/blob/master/t/one-core-peak.sh Vincent
