[Drbd-dev] [PATCH v3 14/16] Gut bio_add_page()

[Kent Overstreet]

On Tue, May 29, 2012 at 11:15:58AM +0900, Tejun Heo wrote:
> Hello,
> 
> On Tue, May 29, 2012 at 12:08:15PM +1000, Dave Chinner wrote:
> > > i thought a bit more about it and the only thing which makes sense to
> > > me is exposing the stripe granularity for striped devices -
> > > ie. something which says "if you go across this boundary, the
> > > performance characteristics including latency might get affected",
> > > which should fit nicely with the rest of topology information.
> > > Martin, adding that shouldn't be difficult, right?
> > 
> > We already have the optimal IO size/alignment field in the topology.
> > Doesn't this fit what you want exactly?
> 
> I don't know how xfs/ext4 is currently benefiting from
> merge_bvec_fn(), so I'm unsure.  If the existing ones are enough,
> great.

Excepting readahead I don't think they are at all.

For readahead all we need is a hint - call it "atomic IO size" or
something. Assuming one of the gazillion things in queue_limits doesn't
already mean that anyways.