Q: Why decreased performance on 2.6.17-rc2-gitX with dual cpu vs single cpu ?

Q: Why decreased performance on 2.6.17-rc2-gitX with dual cpu vs single cpu ?

[Danny ter Haar]

tyan amd64 box with dual opteron 250
4GB ram
debian 64 bit os

It's running as a usenet gateway (2TB in & out roughly each day)

I recently installed munin to track more information.

I run 2.6.17-rc2-git4 in UP mode for 29 hours and then
switched to 2.6.17-rc2-git5 and enabled both cpu's (smp)

irqbalance is installed :


# cat /proc/interrupts
           CPU0       CPU1
  0:      55171    2161605    IO-APIC-edge  timer
  1:          5          2    IO-APIC-edge  i8042
  4:         62        422    IO-APIC-edge  serial
  9:          0          0   IO-APIC-level  acpi
 16:     239941    5682160   IO-APIC-level  aic79xx
 17:    3375173   36339647   IO-APIC-level  aic79xx, eth3
 18:    1433719   67869071   IO-APIC-level  acenic
NMI:       3858       7945
LOC:    2216253    2215748
ERR:          0
MIS:          0


Kernel config's, kern.log output of both kernels can
be found at: http://newsgate.newsserver.nl/kernel/

munin graphics: 
http://stats.handelsweg8.nl/munin/newsserver.nl/newsgate.newsserver.nl.html

I see in the munin graphics some idle time for the first 
time when both cpu's are enabled. But IO-Wait is also much
higher then in UP mode.

Any brilliant mind can explain to me what is happening here ?
i certainly don't get it !

Danny

Re: Q: Why decreased performance on 2.6.17-rc2-gitX with dual cpu vs single cpu ?

[Danny ter Haar]

>munin graphics: 
>http://stats.handelsweg8.nl/munin/newsserver.nl/newsgate.newsserver.nl.html

I forgot one very important thing:

Ethernet stats where you can see less performance than usual:
http://stats.handelsweg8.nl/mrtg/html/newsgate.html

Sorry for that.

Danny