queuedepth_nr_request

https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/performance_tuning_guide/sect-red_hat_enterprise_linux-performance_tuning_guide-storage_and_file_systems-configuration_tools
>For Infiniflash :
The QD needs to bee set to 32 and nr to 16 considering there are 4 active paths to the drive now and aggregate should be <=128.
Each BSSD device total QD should be <=128
# Reads are issued one by one and not affected by QD.
# QD is for writes

#Increasing a disk queue depth might improve disk performance by increasing disk throughput (or I/O) but might also increase latency (response delay). 

#nr_requests: Maximum number of read and write requests that can be queued at one time before the next process requesting a read or write is put to sleep. 
Default value of 128 means 128 read requests and 128 write requests can be queued at once. 
Larger values may increase throughput for workloads writing many small files, smaller values increase throughput with larger I/O operations. 
You could decrease this if you are using latency-sensitive applications, but then you shouldn't be using NOOP if latency is sensitive!
#nr_requests is applied separately for reads and writes within the IO scheduler. So total read+write requests in IO scheduler = 2*nr_requests
#nr_request only applies to the IO scheduler not to the IOs already dispatched to the disk.

Max outstanding limit of IO operations to a device=(2*nr_requests) + queue_depth

#Queue Depth(actual physical disk or Lun queue depth --> avgqu-sz)
[root@localhost ~]# cat /sys/block/sdio/device/queue_depth
32

[root@localhost ~]# cat /sys/block/sdio/queue/nr_requests
16
[root@localhost ~]# cat /sys/block/sdio/queue/scheduler
[noop] deadline cfq
[root@localhost ~]# cat /sys/module/intel_idle/parameters/max_cstate
9