[Orca-users] cpu useage

[Blair Zajac]

Sean O'Neill wrote:

> Shea Martin wrote:
> 
>> I have orca collecting/processing data for over 100 nodes, soon to add 
>> another 32.  It is eating up as much CPU as it can get on 900MHz SPARC 
>> III Cu (usually 100%).  It has been running that way now for 2.5 
>> weeks. Is this to be expected?  Do you guys generally designate a 
>> specific network node to just run orca?
> 
> 
> Some folks even break the servers up into groups and have several 
> machines doing Orca because of the CPU utilization.
> 
>>
>> What happens if I stop the orca script for a few weeks.  Then run it 
>> for a week?  Will there be holes in the data, will the orca script try 
>> to start where it left off, and catch up?
>>
> 
> Assuming the orcallator.se script is still running on the clients 
> collecting data, the orca.pl script will catch up.  It will take a while 
> though considering the number of the machine you have above but it will 
> eventually catch up.  There shouldn't be any holes in the graphs as long 
> as your the orcallator.se text data doesn't holes.
> 
> 
> BTW, this reminds me of something I've been thinking about recently. One 
> thing the bugs me about Orca is that it is SSSSOOOOO CPU intensive. 
>  Really not Orca doing this, its really RRD.

Have you done any profiling of this?  I've got a static version of Perl 
compiled with gcc 3.4.0 with -g3 -O3 -pg to get some numbers but haven't 
done that yet.

> Blair, how hard would it be to put the orca.pl script on each client but 
>  change the code so that it doesn't generate the graphs ... it only 
> updates the RRDs ?

It already takes the flag

   -no-html          Update RRD files and images but not HTML files

so with this flag and with

   -no-images        Update RRD files but not image and HTML files

could get a good way there.

The other thing to do is to dynamically link librrd.so with the SE 
interpreter and have orcallator.se directly create the RRD files.  I 
don't know about this though.  It's nice to have the text files around 
theoretically, but I don't know how many people actually go back to the 
text data files.

 > Then you could play a game with resource manager
> giving the orca.pl script a specific number of CPU shares so that it 
> doesn't overrun the server when something else is going on.  When the 
> box is basically idle, orca.pl gets all the shares.
> 
> Then use ssh/rsync to push  both the orcallator.se text data (for safe 
> keeping if you wanted to that is) and the RRDs.  The Orca server then 
> reads the RRDs and does nothing but updates the graphs and HTML ?

Good idea.  You would also need to push all the orcallator.cfg files out 
so that they remain consistent.

Best,
Blair