[rrd-users] Sar data

Mark Seger Mark.Seger at hp.com
Mon Jul 30 19:47:45 CEST 2007


interesting.  I've never heard of collectd.  I will say it's interesting 
to read that it's 'written in c' to make it faster but then forks 
script.  when I wrote collectl I found it work it to write my own 'cat' 
routine because it was much faster to open/read a file in perl than exec 
cat.  go figure.  9-)
-mark

Ulf Zimmermann wrote:
>> From: rrd-users-bounces at lists.oetiker.ch [mailto:rrd-users-
>> bounces at lists.oetiker.ch] On Behalf Of Mark Seger
>> Wes wrote:
>>     
>>> On 7/30/07 10:37 AM, "Mark Seger" <Mark.Seger at hp.com> wrote:
>>>
>>>
>>>       
>>>> I've been thinking about adding rrd style output to collectl but am
>>>> waiting for someone to raise their hand and say they'd like to work
>>>>         
>> with
>>     
>>>> me.  As for perl being too heavyweight, collectl uses less than
>>>>         
> 0.1% of
>   
>>>> the cpu reporting about 100 different variables.  If you want less
>>>> frequent samples as most rrd users do, you can probably get it down
>>>>         
> to
>   
>> 0.01%
>>     
>>> I figured programs collecting real-time sar data in RRD would be all
>>>       
>> over
>>     
>>> the place.  I was very surprised to find such a thing so scarce.
>>>
>>> I am looking at colleting 5-10 second samples.  I may have to resign
>>>       
>> myself
>>     
>>> to using perl for the sar data.  I have a number of other
>>>       
> statistics,
>   
>> some
>>     
>>> of which have to be gathered by scanning very large log files in
>>>       
> real
>   
>> time.
>>     
>>> By using C, I had hoped to save as many cycles as possible, and be
>>>       
> able
>   
>> to
>>     
>>> use a common set of functions to transmit the data to the central
>>>       
>> collection
>>     
>>> system.  I guess I can always roll my own.  We have times when every
>>>       
>> cycle
>>     
>>> we can save is crucial (can't just up and replace the systems with
>>> faster...)
>>>
>>>       
>> 5-10 second samples is what collectl is all about and it can collect a
>> lot more data than sar so perhaps at least some of what you're looking
>> for it may already generate.  it's not unusual to have collectl
>>     
> running
>   
>> on systems sustaining >90% cpu loads
>> -mark
>>     
>>> Wes
>>>       
>
> I am using collectd on close to 100 nodes, specific per cpu stats, load,
> memory, swap, processes, interface, disk space used and disk IO (ops and
> bytes). It does other things too, but I am not interested in them, so I
> only load those plug ins. Choose collectd for the reason of 10 second
> interval samples.
>
> I have written a grapher for our use but as collectd 4.x splitted the
> structure how RRDs are saved, it won't work for it. Currently planning
> on writing a new grapher soon, one which can hopefully also be released
> as opensource.
>
>   



More information about the rrd-users mailing list