[Orca-users] Orca Speed

[bias16]

What is the typical delay in getting graphs generated for orca?

I was experiencing nearly a 3 hour dealy in graphs, and made the
following changes which have reduced delay to 90 minutes:

* ~ 130 servers
* Generate daily, weekly, monthly graphs only
* upgraded to orca 0..27b
* run orca continuously as daemon
* installed RichSE 3.2 everywhere
* using 1.28 orcallator.se
* using 1.28 orcallator.cfg
* client-based rsync to server every 5 minutes
* orcallator/rrd is on 2 disk stripe
* T1400 with 2 450Mhz CPU and 2Gig memory

I would like to see a delay of less than 30 minutes on all graphs.
Is this realistic?

I have the following changed/upgrades scheduled in coming week:

* Add 2 more CPU's
* Increase memory to 4Gig
* substitute internal disk writes with dual path fibre (EMC) drives
* also considered lowers the number/time of rsyncs but that will
  require more work/planning.

I initially lowered the number of servers in anticipation of running
data crunching on two servers, but wonder how low I must go.

I can also purchase a Sunfire 280 for this processing if the new
arch and dual 900Mhz would greatly help.

I was wondering if the following would help and if anyone has tried
them:

1) Run multiple orca perl script processing.  It looks like script 
it bound to single CPU so not sure extra CPU's will help a lot.  I
figure I could run initial orca on all data, so I can generate all
the needed html.  I would then STOP that and run multiple instance
of orca to generate PNG only.  It looks like I would diff cfg
(specifying difference file reg match) and diff lock... but would like
to put the png in that same data structure.  I really enjoy comparing
the parameter graphs for all systems in single screen.

2) Put rrd data in /tmp.  I'll have 4Gig of memory, so it seems like I
would have plenty of space to attempt this.  Orcallator and html data
needs to survive reboot, but I don't see why rrd data can not be
re-generated.  I'm figuring this would help performance/speed in
generating graphs.

Any other suggestion welcomed.

Thanks,
Liston