Re: SMP on SS10 with 2.4
Hi Emanuel.
I will keep track on this problem, but I have another problem ...
At the moment, my SS20 is my main internet machine. If I kill it, my email is
gone. I am interested in a solution, too, but first I need another machine
doing my email and internet.
I hope to receive missing parts for my new Ultra1 (keybord, graphics) this
weekend. If so, I will start to move my internet account to that machine. It
will take some days, but if that works, I will start testing on my SS20.
So: Please be patiant :-)
Hartwig
On Friday 29 November 2002 18:14, you wrote:
> Hi again
>
> I've done some more experiments:
> I have two 64meg and two 32meg modules in my ss10.
>
> If I put in only one module, 2.4.20rc2 works fine.
>
> If I put in the two 64meg modules, 2.4.20rc2 works nearly well. First the
> modules have to be in the right slots (see sun homepage). Second there are
> some moments the system sleeps for about 10 seconds. I could not yet
> determine what causes this stops besides the fact, that it is a
> memory-related problem (what a surprise ;) )
>
> If I put in all four modules, I get where I was before - nothing works.
>
> Besides this I noticed some strange differences between 2.4.20rc2 and
> 2.2.21 (both smp):
> - With two 64meg modules, 2.2.21 counts about 127meg, 2.4.20rc2 counts
> about 124meg. I did not know that the kernel version does have an affect
> on the installed memory ;)
> - With 2.2.21 the dual hyperSparc at 125Mhz produces about 125bogomips per
> cpu, with 2.4.20rc2 it produces about 109bogomips. This decrease in
> doing
> nothing has been observed before on suse-sparc.
> If I calculate 2^100000 with bc, there is no time difference for the two
> kernels. This probably does not mean anything, I just wanted to see if I
> can prove this in an easy way.
>
> Can someone reproduce similar problems (Hartwig?)?
> Is there a sparc-kernel-mailing-list or is lkml the place for such
> problems?
> Did I something that I should not on sparc? I'm sparc-newbie...
>
> Thanks
> Emanuel
>
Reply to: