A: Linux SMP is slower than Solaris, FreeBSD, but better/worse than NT
etc..
B: That was 2.0 SMP. 2.1 is a lot better I hear...
Thats pretty vague, if not wrong (I'm no expert). And I know its hard
to give a single number to "SMP performance" :) Still, I'd like to know
something about how much overhead we incur. I guess the kernel support
has improved cuz its not all under one big lock anymore (2.1)... is
there really a big difference between 2.0 & 2.1? How much should
overhead decrease when people start using spinlocks? And how much will
the 'coolness factor' increase? Maybe by roughly a factor of 1.5?
BTW, There was a big flurry of news about I2O a while back...Is this
really a big deal? It seems that Ingo's IO-APIC patch allows all
processors to take interrupts (is that NEW?)... And I think I heard that
I2O is merely some spec that allows you to have separate I/O processors
beside the main CPUs. I think something like this was used in the Intel
Paragon MPP (NON-SMP) supercomputer... but APIC seems a lot more
important to most people :)
Thanks for any info you might route my way :)
-BenRI