[lkml]   [2002]   [Jan]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] O_DIRECT with hardware blocksize alignment

Wouldn't this cause, single IO to split into (possibly) 3 I/Os for
RAW VARY on all unaligned IOs ? In that case, wouldn't it be better
of doing entire I/O with 512byte buffer heads, so that all of them
can be merged ?

IMHO, preventing the merge due to b_size mismatch just to handle few
drivers/hardware may be overkill. This will penalize all the drivers
which can handle variable size IOs also. Isn't it ? Leaving it to
the driver and doing variable size I/O on only the drivers that can
handle them, may be a better option ?

In my initial version of RAW VARY patch, I split the I/Os into possibly
3 I/O in raw.c. Do brw_kiovec(.., 512) for the first page and last pages &
4K for the middle pages. It showed reduction in CPU utilization for
large IOs. But it since it could split single I/O into 3 I/Os, i did
not see significant I/O throughput improvement.


> On Tue, Jan 15 2002, Jens Axboe wrote:
> > On Tue, Jan 15 2002, Andrea Arcangeli wrote:
> > > actually we could also forbid merging at the ll_rw_block layer if b_size
> > > is not equal, maybe that's the simpler solution to that problem after
> > > all, merging between kiovec I/O and buffered I/O probably doesn't
> > > matter.
> >
> > Agreed, this is also what I suggested.
> Here's the right version, sorry. This still potentially decrements
> elevator sequence wrongly for a missed front merge, but that's an issue
> I can definitely live with :-)
> ....
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:23    [W:0.050 / U:4.908 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site