Hello, this patch series contains one tiny cleanup and two relatively simple fixes for BFQ I've uncovered when analyzing behavior of four parallel sequential readers on one machine. The fio jobfile is like: [global] direct=0 ioengine=sync invalidate=1 size=16g rw=read bs=4k [reader] numjobs=4 directory=/mnt The first patch fixes a problem due to which the four bfq queues were getting merged without a reason. The third patch fixes a problem where we were unfairly raising bfq queue think time (leading to clearing of short_ttime and subsequent reduction in throughput). What do people think about these? Honza
