Messages in this thread | | | Date | Sat, 28 Nov 1998 15:27:18 +1300 | From | Chris Wedgwood <> | Subject | Re: SMP scalability: 8 -> 32 CPUs |
| |
On Fri, Nov 27, 1998 at 08:38:41PM -0500, Dave Wreski wrote:
> 1. Because there aren't many PCs with 8 or more CPUs around to do > testing on. > 2. Because of #1, our efforts are better spent working on more > real-life problems. > > Among more techincal reasons, I'm sure.
Traditional SMP doesn't scale well above a certain threshold either.
In a previous life I tested the scalability of SGI Onyx machines and there is a real performance/processor trail off above about 8 processors -- although I'm sure this number is application specific, 8 was about right for the code I was testing with.
I think the main reason is memory bus contention and the fact you have many many caches to keep coherent.
ccNUMA (cache coherent Non Uniform Memory Access) architectures like the Onyx II and Origin scale considerable better, to hundred of processors for some applications. Neat stuff if you have the money.
-cw
- 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/
| |