[Xymon] Fwd: About CPU load

[Steve Holmes]

We also have regular expected load spikes (e.g. When backups are running) and handle those with the TIME keyword on the LOAD line. 
Steve


Wherever you go, there you are.  

On Apr 8, 2013, at 8:21 AM, Ralph Mitchell <ralphmitchell at gmail.com> wrote:

> You can handle load spikes with the DURATION keyword in alerts.cfg.  The client normally samples every 5 minutes, so if a spike generally lasts for two samples you could set DURATION>13 to get alerts sent out on the 3rd sample.
> 
> Ralph Mitchell
> 
> 
> 
> On Mon, Apr 8, 2013 at 4:36 AM, Iain M Conochie <iain at shihad.org> wrote:
>> On 2013-04-07 17:22, Operaciones wrote:
>>> Thank you Ryan, one ask, are there eny way for calc this?, or you
>>> suggest taht the better way is testing in each server?
>> 
>> Load average is a logarithmic method of measuring the run queue on your system. The output of the 3 values gives you the run queue over 1 minute, 5 minutes and 15 minutes respectively.
>> 
>> An old rule of thumb was to have a load average of not greater than 2 per cpu, but this is generally site dependent. Are you going to be sending out alerts for a high load? Do you _really_ want to be woken at 2AM cos the backup job has spiked the load?
>> 
>> I would advise you set the threshold a bit higher for your system (perhaps 8.0 and 16.0, based on 4 cores) and then after a suitable time period (perhaps 2 weeks) you can look at the CPU load graph to see what kind of usage the system has and adjust accordingly. During this time you probably don't want to send out alerts (but perhaps you do; your call :) IMO, normally a spike in load is not an issue, but a a system that over 6 months has an ever increasing load is a problem. This is where the trends analysis function comes into it's own.
>> 
>> OF course load spikes can impact performance, but it is hard to gauge exactly what that impact is from the load average alone.
>> 
>> Cheers
>> 
>> Iain
>> 
>> 
>>> 
>>> El 07/04/13 18:16, Novosielski, Ryan escribió:
>>> 
>>>> This default is a reasonably nice default for a single CPU system,
>>>> but this is really a parameter that is site-dependent. For example,
>>>> I have a 24 CPU compute server. A load of 5 is hardly breaking a
>>>> sweat. In some cases, high load doesn't even affect the system and
>>>> you wouldn't want to be notified until a normally unreasonably high
>>>> value. I'd suspect 10.0 15.0 for your system or 15.0 20.0 might not
>>>> be a bad guess.
>>>> 
>>>>  
>>>> 
>>>> FROM: Operaciones [mailto:operaciones at corpresa.com [1]]
>>>> SENT: Sunday, April 07, 2013 10:58 AM
>>>> TO: Xymon List <xymon at xymon.com> [2]
>>>> SUBJECT: [Xymon] Fwd: About CPU load
>>>> 
>>>>  
>>>> I forget, the default params that appear in this file are:
>>>> 
>>>> LOAD    5.0 10.0
>>>> 
>>>> Thank you, best regards.
>>>> 
>>>> -------- Mensaje original --------
>>>> 
>>>> ASUNTO:
>>>> About CPU load
>>>> 
>>>> FECHA:
>>>> 
>>>> Sun, 07 Apr 2013 16:56:32 +0200
>>>> 
>>>> DE:
>>>> Operaciones <operaciones at corpresa.com> [3]
>>>> 
>>>> PARA:
>>>> xymon at xymon.com [4]
>>>> 
>>>> 
>>>> Hello,
>>>> 
>>>> I'm using xymon 4.3.5, my as is the follow:
>>>> 
>>>> Is correct the default params that appear on "analysis.cfg" for
>>>> chect the cpu load in case that the server have 4 cores (intel xeon
>>>> 1230)?, can i use this values for anyone CPU types?
>>>> 
>>>> Thank you, best regards.
>>> 
>>> 
>>> 
>>> Links:
>>> ------
>>> [1] mailto:operaciones at corpresa.com
>>> [2] mailto:xymon at xymon.com
>>> [3] mailto:operaciones at corpresa.com
>>> [4] mailto:xymon at xymon.com
>> 
>> _______________________________________________
>> Xymon mailing list
>> Xymon at xymon.com
>> http://lists.xymon.com/mailman/listinfo/xymon
> 
> _______________________________________________
> Xymon mailing list
> Xymon at xymon.com
> http://lists.xymon.com/mailman/listinfo/xymon
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.xymon.com/pipermail/xymon/attachments/20130408/b9f19dd5/attachment.html>