Re: SMP scalability: 8 -> 32 CPUs

Henrik Olsen (henrik@iaeste.dk)
Tue, 1 Dec 1998 03:41:39 +0000 ( )


On Tue, 1 Dec 1998, Chris Wedgwood wrote:

> On Mon, Nov 30, 1998 at 10:38:22AM -0500, The Galaxy Ranger wrote:
>
> > With all this talk of Linux not scaling past 8 CPUs, does this
> > imply that Linux will never be ported to the connection machine? It
> > is Sparc based after all. Speaking of which.... Sun borrowed that
> > technology when designing the E10000. Does that mean if you load
> > Linux up on that system you will only be able to use 8 of the 64
> > CPUs?
>
> Linux runs on the AP1000+. (Check arch/sparc/ap1000).
>
> Check out http://cap.anu.edu.au/cap/projects/linux/ for more details.
>From a quick browse of this, it would see like the AP1000+ should be
considered a cluster, not an SMP machine, as it's a message passing
distributed memory thingie, not a shared memory, shared interrupt system
such as the SMP kernets are made for.

These two architectures need radically different things to scale well,
fast message passing/routing for the AP1000+, finegrained resource locking
and efficient interrupt brokering for the SMP system.

The main obstacle for Linux scalability according to Linus is the things
that still depend on the global kernel lock.

BTW, the problem above 8 processors isn't that linux don't use them, but
that the time spend by each of them waiting for one of the others to
release the kernel lock, makes it not performance efficient.

-- 
Henrik Olsen,  Dawn Solutions I/S
URL=http://www.iaeste.dk/~henrik/
Get the rest there.

- 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/