[Orca-users] Re[2]: How many orca installations

[Rusty Carruth]

duclosd at post.ch wrote:
> 
> Hi again,

(btw - have you joined the list yet so we can avoid ccing you?  Just curouis)

> 	Allen Eastwood [mixal at swbell.net] wrote:
> > I'm not sure what your point is about running something on every host.
> > With any performance collection tool, and I've had to evaluate several,
> > you have to run some sort of collection process on each machine for
> > which you want data.
> > 
> Well, in the MRTG setup I have now, one machine ssh's to the others every 5 minutes
> to collect the data, there isn't anything installed on the remote hosts. But then I don't
> have all hosts setup yet and I do wonder how much load this will create on the MRTG
> server and on the network when it's querying 60 hosts every 5 minutes about 8-10
> different performance criteria. 

Think about my setup.  I'm running NFS across a T1 link, with the timeouts
set WRONG so that I often time out and request the data a 2nd time while
the first 8k is on its way.  I'm monitoring 15 hosts, ALL values possible,
(PLUS the new State values, just added in the last month).  If you used
rsync you'd never notice the load, since its only a line of data that changes
every 5 minutes!  But then, I'm talking about the raw data files...

> What I like is that there are no dependencies that bind
> the monitored hosts. They don't mount anything, transfer anything or otherwise do
> anything that could go wrong and affect their functioning. 

The 15 machines I'm monitoring I don't even have root access to.
If you only update the information every 5 minutes the load is truly
negligible, as far as I can tell.
(On the root issue - if none of the things you want to monitor require root access,
you might consider running it as a lowly user like I'm doing.  Should give you
even warmer fuzzies ;-)

> What looks nice about orca is the way it displays the data according to host or data type,
> so you can compare in different ways (between hosts, between performance criteria).

Plus with the new State graph you can see at a glance the entire state of your
machine.

> > ...
> How heavy is SE toolkit? 

In my experience, nobody will notice.

> Does it install cleanly? 

What version of Solaris?  What version of the toolkit?  On the OLD solaris
we use (2.5.1) I had to use the old SE toolkit, and it had absolutely
no problems.  I cannot speak for the newer versions.

> These are in large part productive
> systems where I can't risk negative effects on the application. They are also managed
> by a large number of people and it's difficult to ensure homogeneous setups.
> How heavy is the graph generation?

I'd NOT do the graph generation on the remote machines - at least not until
you get the first set done (i.e. up to current).  With the 15 machines I'm
monitoring, removing all the files and forcing orca to completely start
from scratch, my machine is very busy for a long time (on the order of
hours, as I recall).  Of course, it IS only cpu load, so interactive
performance is not affected much.  YMMV ;-)

> > To make things easier, I would recommend that you you make a package
> > with the SE toolkit and Orca and the configuration you want.  We also
> > use this and it works great, especially since we have it as part of our
> > custom jumpstart.
> > 
> Yes, that sounds like a good recommendation. Then I guess it's rather up to me whether
> the installation is clean. :-)  I think I'll try this on some test systems. Does it take a
> while to get all this setup? Or do the sample config files pretty much do what you need?

I'd say the sample config files cover just about everything you need -
it took me more time to figure out where on my machine I was going to put the files
than it did to set up the config file once I'd decided that little issue!

Of course, I installed it in the way it expected ;-)  (one shared directory where
each host has a subdir holding its orcallator files (that's the NFS mounted dir
I've mentioned before), and 2 local directories where the rrd files and the html
files go).

rc