[Orca-users] (no subject)

[Liston Bias]

On Tue, 10 Dec 2002, Sean O'Neill wrote:

> >We actually use PUSH.  It seems to be much better on the performance side
> >and I didn't want orca having to keep track of systems that happen to up.
>
> Curious what you mean here by better performance using a PUSH model vs a
> PULL model. One strong distinction is the data "gets there faster".

Faster because I had 250 threads running at once.  When the clients rsync
in, I have all the data updated in a matter of seconds.  When I was doing
the PULL model through single script, it would run the rsyncs sequentially
so it would take many multiple minutes to update the data.  If a
client/network was hung then I would fall behind more than my 5 minute
window.  I suppose I could spawn off processes but that could add
problems.

The fact that some of my clients & networks go down periodically added to
the performance problems of PULL.

In general, we like for our clients to be responsible for what is
happening to/on them.  We have many PUSH scripts that run on clients.
And maintain configuration information through client-based cfengine.

> Another reason I did a PULL config is I have a Ultra5 acting as the Orca
> server.  Its already pegged at 100% utilization 100% of the time.  By using
> a PULL, I limit the impact of data transfer on CPU utilization.  I would
> rather the CPU be crunching RRD and graphs then 50 rsyncs.

Orca is gonna pretty-much peg anything unless the graphs as staying less
than 5 minutes old.  I'm mostly concerned with how orca keeps up with
graphs.  I didn't see a difference with the PULL/PUSH scenarious, but you
definitely get more control over the load comes in.

I'm currently running orca on 480R so may architecture could have impact
on what does and does not have an impact.

- Liston