Thank you Florian for the update. In my case, only 2 or 3 servers out of 10 are showing the discrepancy. The hardware, OS, collectd and related versions are same. Please see the attached images.<br><br>--<br>Shib<br><br><br>
<div class="gmail_quote">On Tue, Jun 3, 2008 at 11:17 PM, Florian Forster <<a href="mailto:octo@verplant.org">octo@verplant.org</a>> wrote:<br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Hi Shibashish,<br>
<div class="Ih2E3d"><br>
On Tue, Jun 03, 2008 at 12:49:29PM +0530, Shibashish wrote:<br>
> In the CPU graphs, in some (2) of the hosts, the vertical axis starts as 0,<br>
> 200, 400, 600, 800. And one of the host shows as 0, 0.5k, 1.0k. Why is this<br>
> so ? How can percent go beyond 100 ? Or is the plotting showing correct data<br>
> ? Is there a way to correct this ?<br>
<br>
</div>the CPU plugin collects ``jiffies'' or ``system timer interrupts''.<br>
Typical values for a Linux kernel are 1/100s and 1/250s, but you can set<br>
this to 1/1000s for (near) realtime systems, too.<br>
<br>
So there are 100, 250, or 1000 jiffies in one second. So given the<br>
values I'd guess that the kernel of your distribution is built with the<br>
HZ variable set to 1000 at build time. Unfortunatly I don't know how you<br>
can double check this, but I'm sure Google will be able to assist you ;)<br>
<br>
You can convert this number to a percentage, which is easier to compare<br>
between machines, using RRDTool's ``RPN'' functionality. You'd need to<br>
do something along these lines:<br>
sum=user,syst,+,idle,+<br>
user_pct=user,sum,/,100,*<br>
syst_pct=syst,sum,/,100,*<br>
idle_pct=idle,sum,/,100,*<br>
Then graph `user_pct', `syst_pct', and `idle_pct'.<br>
<br>
Regards,<br>
-octo<br>
<font color="#888888">--<br>
Florian octo Forster<br>
Hacker in training<br>
GnuPG: 0x91523C3D<br>
<a href="http://verplant.org/" target="_blank">http://verplant.org/</a><br>
</font><br><br></blockquote></div>