> On Mon, Feb 19, 2001 at 05:40:09PM +0100, Hubert Feyrer wrote:
> 
>  > > Err.. you'll note that with UBC, it does significantly fewer I/O
>  > > operations...
>  > 
>  > shouldn't this result in *less* CPU use (as no IO needs to be scheduled)?
> 
> I/O generally doesn't cause CPU time unless it's non-DMA.

Oh, actually, this is historically not really quite3 true. I've noticed for
years that even with good DMA devices getting higher IOPS has always been
quite linearly related to CPU speed (both on the disk and the main system).