Hello Alan,
I am now running quite nicely your latest stuff..;-)
It feels GREAT.
But the throughput at least on my system is somewhat lower than with my latest
-aa VM (vm_25) kernel 2.4.18-pre8-K3-VM-24-preempt-lock.
I am going to compare "your" version against
2.4.19-pre2
vm_28
read-latency2
O(1)-K3 (the hardest part)
preempt + lock-break (got that but not Ingo's stuff then)
2.4.18.pending ReiserFS stuff
All OOM problems are fixed with vm_28 for me.
I've checked it with and without swap.
With former versions some system tasks (smpppd), kdeinit and desktop processes
(xperfmon++, kpanel, kmail, kalarm, etc.) were falsely killed.
With 2.4.19-pre2-ac2 + pre and without swap (I disabled it before running the
"test" prog) kswapd (?) goes into 20~25% system time usage and the whole
system gets unusable. I had to reboot...
X have to be reniced -10 to get smooth mouse movement under both kernels
(Ingo's part) during compilations (bzlilo).
Regards,
Dieter
BTW Some numbers will follow when I get O(1) + preempt going with 2.4.19-pre2
+ vm_28.
> All OOM problems are fixed with vm_28 for me.
Excellent - thats one of the important ones.
> I've checked it with and without swap.
> With former versions some system tasks (smpppd), kdeinit and desktop processes
> (xperfmon++, kpanel, kmail, kalarm, etc.) were falsely killed.
As an aside, with the address space accounting code Im testing we can finally
do precise OOM handling.
> With 2.4.19-pre2-ac2 + pre and without swap (I disabled it before running the
> "test" prog) kswapd (?) goes into 20~25% system time usage and the whole
> system gets unusable. I had to reboot...
I've not really done much testing without swap I must admit
Alan