> > And an ia32 process isn't going to run on a zArch processor or vice
> > versa.  I'm pretty sure OpenMosix requires homogeneous processors in its
> > resource blob.
>
> Hmm, I see.
> Well, I don't know what you mean by "homogeneous". If you're referring to
architecture then I think you're right, but the FAQ on the site says that
you
> can have a mix of different capacity IA-32 processors.

All IA32 processors use the same instruction set, however -- for example
IA32 by Intel and IA32 by AMD are the same instruction set.  390 and zSeries
are not in that family of processors and use a different instruction set, so
the binaries would not be executable if migrated to a node with a different
processor architecture.  A simpler example would be if you introduce a IA64
box into the cluster; it's a different type of CPU, and would likely not
work.

> Well, and do you see major obstacles to porting openMosix to z/Linux?

It would probably work as is, but it really doesn't do much towards solving
the problem you described if you want to exploit the Intel systems from the
zSeries. I'm compiling it now to see how bad it would be, but ... don't
expect miracles.

> > Anyone else remember Amoeba?  OK, anyone else *other* than Dr. Boyes.
> > Let's not always see the same hands here.
> I did read about it, a long time ago. But that looks rethorical :) Do you
have a story to tell? :)

Hey, I resemble that remark! 8-)  *Somebody* has to remember this stuff, or
we'll end up reinventing it AGAIN.

Actually, Amoeba goes far beyond what OpenMosix does, but does it by dint of
having a dramatically different internal architecture. *Everything* in Amoba
is mutable and the relationships between nodes are determined by
message-passing, so it really DOESN'T matter where things are. It's a
standard Tannebaum "elegant is cool" design. Too bad he got bored with it --
it's a fascinating system. Scales REALLY well, too.

Reply via email to