Linux 5.0 HDD I/O Scheduler Benchmarks - BFQ Takes The Cake

FS-Mark with four threads saw Mq-deadline and BFQ low-latency slightly lower than the others on the consumer WD Green drive but not much variation between the other options or with the higher performing VelociRaptor.

In another FS-Mark environment, BFQ low-latency was much slower than the rest for the WD5000AZRX and MQ-deadline also was on the slow side. But when running the WD1500HLHX with Mq-deadline, it came out ahead of BFQ and the others.

BFQ came in slightly lower than the other choices for the compile bench workload while Mq-deadline picked up two victories.

In the more processor-focused workload of compiling the Linux kernel, the I/O scheduler wasn't a bottleneck at all in the compile performance for this real-world workload on the two consumer SATA 3.0 HDDs.

BFQ doesn't perform well for PostgreSQL database server workloads.

When looking at the geometric mean of all the benchmarks carried out, BFQ came out in front on both drives. BFQ performed very well particularly in the responsiveness tests of still quickly starting up applications when there was I/O happening in the background and generally performed well in the other tests too. In some of the heavy throughput tests though, BFQ fell behind. BFQ might not be the best choice for servers, but for desktop systems its low-latency mode performs very well both for consumer SSDs and HDDs. Though if you are tweaking your I/O scheduler to enhance the responsiveness of your system, it may be time to consider switching over to solid-state storage considering the falling prices even on NVMe SSDs these days and offering much more performance potential than what can be obtained simply from tweaking your kernel.