hi,

we already have code to do i/o clustering, and I've verified that
it works for sequential i/o.  (or at least it worked when I tried it
back before UBC was merged.)  it appears that it's not working now
for the i/o pattern that tar extraction uses, though.

-Chuck


On Thu, Jan 04, 2001 at 06:23:27AM +0100, Reinoud Zandijk wrote:
> Hi Chuck,
> 
> On Wed, 3 Jan 2001, Chuck Silvers wrote:
> > yea, there is some performance work remaining to be done for softdeps
> > in the UBC world.  (and other areas too, but this is one of the more
> > noticable ones.)  I haven't had a chance to investigate yet,
> > but it's on my list of things to do.
> 
> Well... if i look at IWAMOTO Toshihiro wrote, would it be an easy fix to
> at least glue individual pages that need to be written out together? so
> continuous pieces of memory and disc can easily be glued as one operation
> and thus not hitting the maximum number of transactions for a
> device? maybe in a much lower level of disc scheduling perhaps? as a more
> general issue? havent looked at code yet, but maybe its more general? or
> even implemented allready ?
> 
> just my $0.02c :)
> 
> Cheers,
> Reinoud