On Mon, Mar 06, 2000 at 09:50:39AM +0100, Rogier Wolff wrote:
> Matti Aarnio wrote:
> > On the contrary, those vector systems do have a MMU, but it isn't PAGED.
> > And similarly, there is no demand PAGING, there is only SWAPPING.
> > Everything of process out of memory, everything in to the memory.
> > (This is one reason why those machines have humongous IO bandwidth..)
>
> Aaargh!
>
> Does that mean that I can't run:
...
> with say 1.5G datasets and just 4G of RAM?
>
> 4 * 1.5G is 6G, so it doesn't all fit into RAM at the same time, while
> every processing step uses only 2 * 1.5G = 3G of RAM, which fits
> easily.
On those vector CRAYs there is a requirement that your
memory usage won't exceed the available main memory.
That is, it isn't any sort of VIRTUAL MEMORY beast.
(And while your vector performance is tens or hundreds
of gigaflops, you don't want to wait for pageing to
mess around for 5-20 milliseconds for your precious
data page -- not that vector subsystem state is so
easily storable/restartable at bus-address fault either...)
On CRAYs the king of the design are those vector gigaflops.
*Everything* else is scrapped if it isn't helping there.
( -> address fault during vector execution kills the instruction
without a change for restart. )
> Roger.
> ** R.E.Wolff@BitWizard.nl ** http://www.BitWizard.nl/ ** +31-15-2137555 **
/Matti Aarnio
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Mar 07 2000 - 21:00:19 EST