Re: Asymmetric multiprocessing?

Rik van Riel (H.H.vanRiel@phys.uu.nl)
Sun, 3 May 1998 17:30:03 +0200 (MET DST)

On Sun, 3 May 1998, Rogier Wolff wrote:
> Alan Cox wrote:
> >
> > > For above configuration I'd say ignore the '020. It just produces very
> > > serious congestion of the memory bus on the motherboard.
> >
> > Or use it as a glorified DMA controller/IRQ buffer
>
> This popped up yesterday: if you want to build a large cluster of
> machines with humungous amounts of processing power (beowulf), would you
> put one or two processors in each node?
>
> Now as an alternative I suggested putting one REAL processor on each
> node (P-II/400) and one that does all the nasty side-tasks (say a
> P-II/200)
>
> That should allow the compute-intensive task to run at 100% CPU (not
> just 98%) leaving all the interrupts and other stuff to the other
> processor.

Buzzword: Transputer.
Mainframe era buzzword: frontend computer.

ie: You run the operating system on the 'main' (OK,
the right word is central) computer. Then the OS
runs the app(s) on the other CPU+RAM boards on the
BUS.

IE:
- 1 Pentium 166 + 128M + loads of disks + network
- 3 PCI cards with single PII/400 (or a specialized
DSP for that matter) and 32 MB RAM.

Then the programs need to be split up in 'small' parts
and the OSes can give each fast CPU a task to chew on.

Somewhat like the push equivalent of distributed.net :)

Rik.
+-------------------------------------------+--------------------------+
| Linux: - LinuxHQ MM-patches page | Scouting webmaster |
| - kswapd ask-him & complain-to guy | Vries cubscout leader |
| http://www.phys.uu.nl/~riel/ | <H.H.vanRiel@phys.uu.nl> |
+-------------------------------------------+--------------------------+

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu