After several test, I find that CPU stat info are not correct with kernel 2.6 inside uml.
tested with skas3/skas0 mode (Proc P4HT/PD/Core2Duo):
host kernel 32bit 2.6.20smp skaks3 | guest 32bit uml 2.6.13/2.6.21 skas3/skas0
host kernel 32bit 2.6.14smp skaks3 | guest 32bit uml 2.6.13/2.6.21 skas3
host kernel 32bit 2.6.5 | guest 32bit uml 2.6.21 skas0

TEST1: 1 uml
Commande used in uml : ./stress --cpu 1 --timeout 60s&
TOP in host said:
umluser1 ~%100
TOP in uml said:
stress  ~%90
Value=%90, if config host kernel Timer frequency = 1000 HZ
If Timer frequency = 250 HZ, value=%50

TEST2: 2 umls
Commande used in umls : ./stress --cpu 1 --timeout 60s&
TOP in host said:
umluser1 ~%50
umluser2 ~%50
TOP in uml 1 said:
stress  ~%45
TOP in uml 2 said:
stress  ~%45

The problem is if you monitoring CPU inside UMl, with MRTG for example, the results are not correct and
decrease proportionally with the number of umls running and using CPU at the same time.
With kernel 2.4 the values are correct and Work perfectly with (proc P4HT/PD):
Tested with host kernel 32bit 2.4.x-smp skas3 | guest 32bit uml 2.6.13/2.6.21 skas3

Ne gardez plus qu'une seule adresse mail ! Copiez vos mails vers Yahoo! Mail