Blk-mq/SCSI-mq Are Looking Real Good For Linux Disk Performance

It's been a while since last having any major news to report on the multi-queue block layer for the Linux kernel, but that blk-mq and more recent scsi-mq work is progressing well.

After entering the mainline kernel and then being revised a few cycles, blk-mq is effectively feature complete and fast while more recently, SCSI-mq entered the mainline kernel. The multi-queue block layer can lead to better drive performance through reducing latency by balancing I/O workload across multiple CPU cores and allowing for multiple hardware queues.

For those wishing to learn more on the blk-mq/scsi-mq current work, Christoph Hellwig presented at LinuxCon Europe last month regarding this multi-queue block layer work. The slides are available in PDF form.

The performance improvements shared in the presentation look very good for large systems, Linux 3.18 brings a compile-time option for being able to switch between blk-mq and the legacy code, and in the near term the developers hope to replace the old SCSI I/O code with the blk-mq code. Missing features right now are I/O scheduler support in blk-mq and mainline multi-path support. Check out the aforelinked slides if you wish to learn more about this major, modern improvement to the Linux kernel.