Mailinglist Archive: opensuse-amd64 (105 mails)

["Daniel Persson" <daniel@xxxxxxxxxxx>]

Hi Andreas,

> The numbers are calculated since system start.  The load is evenly
> distributed soon but not in the way that the same interrupte, e.g. for
> eth0 is handled by both cpu0 and cpu1 but in the way that CPU0 and CPU1
> get in sum the same number of interrupts.  So, cpu0 will now handle e.g.
> timer, acpi, eth0 and cpu1 will handle ide0, ide1 and timer.  This also
> might give besser locality.
Ok, thanks for the explanation.

>> One question, im not near the machines right now, but, if they where
>> deliverd with all memory installed on CPU0, and no modules at CPU1,
>> could this generate this kind of problems ?
>
> It shouldn't ;-).
Ok.

>For performance reasons I would change it.
Yes, the right way should be to have the same amount of modules for each
CPU, and if only using 2GB, then one should use 4x512Mb, two per memory
slot - is this correct - performance wise ?
I belive that they are configured this way - is there any way for me to
check this trough the OS on distance.(since im not near the machines)

What is the status on the NUMA code, would it be wise to start using that
instead of the SMP kernels at this stage ?

Since i (should) use the same config on both setups, im feeling that there
most be some serious config mistake somewhere, i mean, 10times slower on
setup 2 is a little to much...

Thanks for all great help so far, i guess that i will have to dive in my
configurations again.

/Daniel