Optimizations have been prepared for the Linux kernel to improve the performance of I/O schedulers

[or1k]

Jens Exbo, creator of io_uring and the CFQ, Deadline and Noop I/O schedulers, has continued his experiments with I/O optimization in the Linux kernel. This time he focused on the BFQ and mq-deadline I/O schedulers, which proved to be a bottleneck at least in the case of high-speed NVMe drives.

As the study showed, one of the key reasons for suboptimal performance of I/O scheduler subsystems was problems with competing locks ("lock contention", an attempt to get a lock held by another thread). Thanks to measures aimed at reducing lock contention, such as serialization of dispatch and query insertion, the speed of the schedulers increased in some situations by times (in IOPS).

When testing the BFQ scheduler with the fio utility, performance increased from 567K to 1551K IOPS, and lock competition decreased from 96% to 30%. In the case of mq-deadline, performance after applying the proposed patches when using NVMe storage increased from 1070K to 2560K input/output operations per second (IOPS), and lock competition decreased from 94% to 23%.