jiho@postal.c-zone.net said:
> On 03-Apr-98 John S. Dyson wrote:
> 
> > FreeBSD does not need to use or create three pages like the old VM code
> > did, and hasn't for the last year++.  The source says it all!!!
> 
> I'm profusely embarassed about this.  I posted a very close follow-up message
> correcting the mistake (which I caught fairly quickly) in no uncertain terms.
>  
If you ever have questions where you are not sure, I am very willing to
privately (or publically when appropriate) answer them as needed.  I
really didn't mean to embarass, this stuff is something that can take
a while to catch up on;  Especially the FreeBSD VM code, where it is
not like anything else.  The published literature sometimes talks about
MACH, or SVR4 schemes or whatever.  We are nearly a third scheme, that
has mitigated most of the MACH problems, by de-genenralizing the code
slightly, but still supporting about everything that a U**X like OS needs.

The object mgmt stuff was fixed by forgetting about the "defer until
absolutely necessary" dogma.  There is still some epsilon (or slightly
larger) overhead in our scheme than could be streamlined out, but
I think that we have demonstrated the feasibility of correcting the
object mgmt problems.  I can imagine that UVM will be able to have
a several percent edge on latency.

Things like zero-copy I/O are architecturally consistant with our
current design, without hackery.  The object and inheritance concepts
do impose on the learning curve, but also are valuable tools once
they are understood.   My goal is that within our framework, we
won't have to "invent" new hacks to gain higher level features, but
mostly we can adapt the already existant framework.

-- 
John                  | Never try to teach a pig to sing,
dyson@freebsd.org     | it just makes you look stupid,
jdyson@nc.com         | and it irritates the pig.