Dennis Clarke wrote:
On 10/10/05, S Destika <[EMAIL PROTECTED]> wrote:
Ok. On the SPEC benchmarks does JBB count as a good one to measure kernel
scalability? If so I posted a reply to barts above - Linux/IA-32 achieves 72%
of perfect scalability when going from 1 to 8 way. Solaris10/AMD64 (all 64bit
kernel, jvm, cpu) achieves 82% when going from 2-way to 4-way. Not apples to
apples but enough to prove the point that Linux scalability isn't bad at all.
I'm a little fuzzy on numbers here but does that mean that a four way
machine will be 28% idle? Which means one processor doing nothing?
Or does it imply that there were application throughput measurments
and the four way result was only 72% of the expected 400% increase
from 1 to 4 CPU's?
Most likely the issue is how well the load is balanced and how much
overhead is involved with distributing the load. An OS that doesn't
scale well will spend a lot of time distributing work, one hit may be
threads moving between CPUs, causing a lot of wasted cache flushing.
Ian
_______________________________________________
opensolaris-discuss mailing list
[email protected]
