Re: Implementation of POSIX shared memory objects

[Mindaugas Rasiukevicius <rmind%netbsd.org@localhost>]

yamt%mwd.biglobe.ne.jp@localhost (YAMAMOTO Takashi) wrote:
> > - Some support from kernel would still be needed, eg. hooks to create and
> >   unlink the object (note that object might stay with the data, even if
> >   there are no references i.e. it's removed only after shm_unlink).
> > 
> > - While some management of objects could be in the userland, I do not
> > really see the point in splitting some parts to userland and some to
> > kernel. Now code is clean and in one place (and it could be a module
> > anyway).
> > 
> > Why? :)
> 
> less kernel code is better unless there are reasons.

It does not seriously increase the complexity of existing code, could be
made a module, and there are better ways to save the space (please see
"Bloat" thread on this list).

> afaik it can be implemented in userland using a directory.  eg. /var/shm
> 

You mean implementing it in on a file-system? While permitted by POSIX,
it would be significantly slower.

> YAMAMOTO Takashi

-- 
Best regards,
Mindaugas