Re: Interesting cross-fertilization with DfBSD

On Wed, 31 Mar 2004, Andre Oppermann wrote:

>        * Avoiding the KVM mappings and the related kernel_object
> manipulations
>       required for those mappings saves a lot of cpu cycles when all
> 
> you
>       want is a quick mapping into KVM.

Yeah -- these issue was discussed extensively at the USENIX ATC FreeBSD
Developer Summit last summer in San Antonio, and also to some extent at
the FreeBSD Developer Summit following BSDCon in San Francisco last fall.
There are tentative plans to move to unmapped and deferred mapping buffers
in our I/O system (Peter, Jeff, Kirk, and Poul-Henning have the details),
but this work is scheduled to be one of the major goals for the 6.x
branch.  Part of our goal, though, in preparing for 5.x has been to try
and make sure APIs are moving in directions to support that goal.  A
simultaneous goal is to untangle the web of connections between the VM
system meta-data and remaining mocked up (duplicated) meta-data for the
buffer cache. 

It will be very interesting to see how it works out for Matt, and where
his proposed approach is similar to or different from the tentative plans
we've discussed.  Obviously, we're willing to learn from any positive
discoveries (and less positive ones) he makes along the way :-).

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Senior Research Scientist, McAfee Research