Alan Cox [off-list ref] writes:

quoted

Anywho, a 40% difference is pretty significant (though NUMA can have
that sort of impact).  Alan, what do you think?  I was never clear on
how exactly the ahci driver would handle a queue depth larger than 32
(if it can't, then clearly we'd need a block driver for this hardware).

The AHCI driver can drop that support in fairly easily all the queue
handling is nicely pluggable.

A 40% improvement is a big win, but it seems to me that if any of that is
not hardware related (ie the bigger command queue) then we need to be
fixing the existing driver to improve *ALL* AHCI devices. Or are we going
to end up with a pile of AHCIish drivers for every random device on the
planet ?

The 40% number was for a queue depth of 32, so yes, there's room for
improvement in AHCI (so long as it isn't related to the block layer's
rq_affinity interacting poorly with NUMA or some such thing).

My request for info on the Micron errata has so far been ignored,
questions on what is involved in the queue stuff likewise. I'd like to
see those matters resolved.

Asai?

-Jeff