> Thus spake Rob Hagopian:
> > That's ridiculuous, I can do the exact same thing today (cat /dev/zero >
> > /tmp/hahawatchthesystemdie ; malloc(1000000000000).
>
> Except that I can enforce a memory usage limit without creating a
> /tmp-space limit, and (although less usefully, I think) vice versa.
>
> > However, you only create the argument for resource limits here.
>
> The argument for resource limits already exists, and is met with
> quotas/ulimit. tmpfs forces me to tie my disk usage and memory usage
> policies together, and I don't think the OS should do that.
Woha! "force"? The OS doesn't force you to do anything. If you _choose_ to
use tmpfs, then yes, I see that quota/ulimit do have problems. Then again,
I always have the problem that they aren't flexible enough to do what I
want anyways. I still think that you make the argument for resource
limits, wether or not they can be handled by quota/ulimit today- I guess
not, so maybe tmpfs isn't a good thing for quota'd systems...
(disclaimer: I don't run quotas on my systems and neither does the CS Dept
which has the Solaris boxes with tmpfs. My reasons are above...)
-Rob H.