[rrd-users] Re: Monitoring CPU stats on Linux? Watch your jiffies.

[Dave Wombat]

So maybe you are seeing a bug in the kernel.  Thanks for the warning.

Generally, one would expect that the change in (user+nice+sys+idle) to
be equal to the total number of jiffies available in an interval.  For
your five minute samples, that would be 30000 jiffies. Well for your
machine it would be 8 times that for the 8 CPUs.  If you sum across
your data lines you don't get 240000.  I see variation of about .1%.

If you want to assume that all the error is in the idle measurement
you could replace idle with (240000 - (user+nice+sys)).  I leave it as
exercise to DEF and RPN that.  If you want to track the error you can
calculate the difference between measured and calculated and put error
bars on you graphs.

--
Unsubscribe mailto:rrd-users-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-users-request at list.ee.ethz.ch?subject=help
Archive     http://lists.ee.ethz.ch/rrd-users
WebAdmin    http://lists.ee.ethz.ch/lsg2.cgi