[cap-talk] Domain change (IPC?) overhead (was: Re: Gnu Hurd status? Power box?)

[Jed Donnelley]

On 3/24/2008 9:52 AM, Jonathan S. Shapiro wrote:
> On Fri, 2008-03-21 at 16:09 -0700, Bill Frantz wrote:
>> nowhere.man at levallois.eu.org (Pierre THIERRY) on Friday, March 21, 2008 wrote:
...
>> One should not get too excited about performance issues. Things
>> like the CPU cost of IPC are not what makes todays interactive
>> applications slow.
> 
> I disagree. Or rather, I agree, but the statement is irrelevant. Since
> today's applications do not use IPC, it is trivially true that IPC is
> not a significant factor in their performance.
> 
> However: robustness implies introducing IPC, and each IPC adds overhead.
> As applications are factored and multiple layers of IPC are introduced,
> performance of IPC becomes critical very quickly.

Isn't the above consideration where the language systems for
POLA provide significant additional value?  That is, if one
can depend on a language system to protect mutually suspicious
code (I'm still a bit skeptical in this area, being reluctant to
accept compilers/interpreters as part of my TCB, but I'm willing to
be convinced), then isn't it possible to avoid IPC overhead and
essentially replace it with subroutine call overhead?

>> For example, today Mark Miller was doing a demo of Causeway, a
>> postmortem causal trace for distributed events. We noticed that he
>> was running it on a Mac (Mach implementing Unix implementing
>> MacOS), running Parallel (which runs Windows on an Intel Mac),
>> running a JVM running Java running E and SWT. The performance was
>> still quite responsive.
> 
> Performance that is acceptable for a demonstration may or may not be
> acceptable for a purchasing contract.

Ain't that the truth.

Regarding "Mach implementing Unix implementing MacOS", I wonder
how much Mach is still in the final MacOS X loads?  This configuration
is not a demonstration system, but is the production configuration
that all Mac systems use.  Does anybody know enough about that
configuration to say what happens when, say, a file read
call is issued?  I guess there is no internal IPC involved.
Can anybody comment with more certain knowledge?

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