[cap-talk] Object-capability vs. monolithic performance

[Jed Donnelley]

At 10:13 PM 1/4/2007, David Hopwood wrote:
>David Hopwood wrote:
> >  - The environment of the driver when running inside the kernel 
> can be made to
> >    look essentially the same as its environment when running outside. Any
> >    particular driver function is always written as though it were 
> running in
> >    a non-kernel domain, even though it may actually be running in 
> the kernel.
>
>In a slightly more cynical (but still serious) vein, this is clearly the
>Right Thing because, if we *disable* this optimization by default:
>
>  - benchmarketeers, other speed-obsessives, and the minute fraction of people
>    who really need to run a GigE card at 95% of the link bandwidth, 
> can turn it
>    on and get the performance they want;
>
>  - the majority of users, including all those who only have a GigE 
> card because
>    one comes with any new PC in 2010, will accept the defaults and 
> get the maximum
>    security and stability;
>
>  - OS researchers will get several papers out of it;
>
>  - everyone is happy :-)

That's were we ended up, though with our "default" switched and I'd say
"content" rather than happy.

--Jed  http://www.webstart.com/jed-signature.html