Re: Linux Kernel Security - Can it ever be 100%
On Fri, Dec 05, 2003 at 11:43:23PM -0500, Joey Hess wrote:
[great stuff which is absolutely correct]
However, I "Tom Ballard" have figured it all out.
The problem with all of computer science is the left hand doesn't know
what the right hand is doing. All of these problems are finite and can
be handled in an "a priori" way. The problem is computer science grew
up not knowing that so we pretend we don't immediately know everything
and compute in an "a posteori way".
What I'm talking about is tearing down the concept of a general purpose
computer. The only reason I can't run all my programs in a single
memory space and know just exactly what the heck is going to happen is
it makes poor economic sense to work that way.
Consider a SQL Server for example. For any given schema which will
a maximum of contain {N1...Nm} records, I can compute "a priori" the
exact disk location of any record. If memory wasn't so fucking slow
and there were plenty of it, we could assemble any image of this very
quickly. All I need is a simple "I/O monster" that does this one fixed
task in an "a priori way".
So the problem is general purpose computers. We need to be able to
produce fixed-function devices in a one-off fashion.
[This rant is probably full of shit] :-)
Reply to: