From your last diagnostics, it show you haven't turn-on turbo write ( md write = auto ), so local disk to disk copy will be slow, if copying was on same disk will much much slow.

 

As mention before, read performance usually was good and easy achieved, I usually use parity check to verify any low-end bottleneck in local first.

 

For example, a Intel J1800 system ( low end 2 core ) with 8 disk reading in same time (party check), the total speed was ~776MB/s and each disk not more then 100MB/s. Bottleneck wasn't CPU nor disks, it is system IO bus bandwidth issue.

 

 

If I simple read a file, i.e "cat /mnt/disk7/xxxxx > /dev/null" then expect result got and also show stable too. The memory (cache memory) full or not won't affect.

 

Edited May 6, 20188 yr by Benson

Thanks for the info but the issue with my system is the inconsistent read transfer rates, that I cannot explain. I think the advice I need is what and how to monitor while performing a read from the array to a Windows machine. What info should I be looking for at the time the big drops are taking place or vice versa when it rises up again to maximum gigabit speed.