The issue he is asking about is why does the md127 device have a
w_await time of 1473.96ms when the highest response time on the
underlying device is 15.94ms.  %util of all devices is low.

Something has to be going on inside the kernel to cause there to be
that sort of a difference.

He is also doing about 3 io/second which is significantly under the
capacity of a spinning disk.

If a disk was having issues with bad sectors/failed sectors/silent
retries then the %util/awaits should show that.

So between when the write gets to the md  device and when the io gets
to the physical device there are 1000+ ms (on average).

And this is a typical sample that is being seen when the array is not
functioning fast, and this issue clears (all by itself) some minutes
or hours later and continues as if nothing happened.